The sensitivity of magnetic particle imaging and fluorine-19 magnetic resonance imaging for cell tracking

Abstract

Magnetic particle imaging (MPI) and fluorine-19 (¹⁹F) MRI produce images which allow for quantification of labeled cells. MPI is an emerging instrument for cell tracking, which is expected to have superior sensitivity compared to ¹⁹F MRI. Our objective is to assess the cellular sensitivity of MPI and ¹⁹F MRI for detection of mesenchymal stem cells (MSC) and breast cancer cells. Cells were labeled with ferucarbotran or perfluoropolyether, for imaging on a preclinical MPI system or 3 Tesla clinical MRI, respectively. Using the same imaging time, as few as 4000 MSC (76 ng iron) and 8000 breast cancer cells (74 ng iron) were reliably detected with MPI, and 256,000 MSC (9.01 × 10^{16 19}F atoms) were detected with ¹⁹F MRI, with SNR > 5. MPI has the potential to be more sensitive than ¹⁹F MRI for cell tracking. In vivo sensitivity with MPI and ¹⁹F MRI was evaluated by imaging MSC that were administered by different routes. In vivo imaging revealed reduced sensitivity compared to ex vivo cell pellets of the same cell number. We attribute reduced MPI and ¹⁹F MRI cell detection in vivo to the effect of cell dispersion among other factors, which are described.

Introduction

Cellular imaging with magnetic particle imaging (MPI) and magnetic resonance imaging (MRI) enables the tracking of cellular therapies and the fate of cancer cells. MPI and Fluorine-19 (¹⁹F) MRI are advantageous as they provide positive image contrast with quantifiable signal, without the use of ionizing radiation. This allows for specific and longitudinal tracking of cells and linear cell quantification. MRI has been used to image therapeutic cells in patients¹, whereas MPI is an emerging modality for cell tracking and currently limited to preclinical studies. In this study, we evaluate the sensitivity of preclinical MPI and ¹⁹F MRI on a clinical system for imaging of two cell types: mesenchymal stem cells (MSC) and breast cancer cells.

Therapeutic MSC have great potential for regenerative medicine. There are over 4000 clinical trials ongoing in the U.S, and over 350 in Canada, for stem cell therapies². After administration, many MSC die due to a hostile, pro-inflammatory environment. Importantly, the number of cells that survive and persistence of cells at the implant site provide information on therapeutic status³. Critical questions about the safety and success of cell therapies—the delivery, numbers, and persistence of cells-remain unanswered. Cellular imaging has the potential to provide answers, and play a role in optimizing dosage, schedules, and administration routes for cell therapies.

Tracking the fate of cancer cells in preclinical models has been another focus of cellular imaging. Metastasis refers to the spread of cancer from the primary tumor to secondary organs and is the leading cause of death for many types of cancer, including breast cancer. The ability to track the fate of cancer populations provides a powerful tool to study metastasis and potential therapeutics which delay or inhibit metastasis. For cell tracking of MSC and cancer cells, imaging quantification is necessary, and so is the detection of few cells (high cellular sensitivity).

Sensitivity for MPI

MPI directly detects superparamagnetic iron oxide nanoparticles (SPIONs), which are used as cell labeling agents. Strong magnetic gradients (T/m) are used to localize SPIONs by creating a field-free region (FFR), and oscillating excitation fields (mT) are applied to alter the magnetization of SPIONs present in the FFR. The FFR is traversed across the imaging field of view and the change in SPION magnetization is detected by a receive coil. The resulting image has positive contrast, and the signal is directly related to the amount of SPION and cell number.

The type of SPION and the amount of SPION taken up by cells are two major factors that determine the sensitivity of MPI for cell tracking. Optimal SPIONs for MPI will strongly magnetize with magnetic fields outside the FFR and experience fast relaxation rates within the FFR. Monodisperse, single core SPIONs with core sizes of ~ 25 nm have been considered ideal^4,5. Cell labeling by SPIONs is typically conducted through co-culture and is dependent on endocytosis. Carbohydrate coatings, such as carboxydextran, increase the interactions of SPIONs with cell membranes, similarly, transfection agents can be used to coat SPIONs to enhance their incorporation to cells.

The most commonly used SPION for MPI is ferucarbotran (VivoTrax™, Magnetic Insight Inc., Alameda, USA), which is repurposed from the original use as an MRI contrast agent (Resovist®, Bayers Healthcare). Ferucarbotran has a carboxydextran coat and is a polydisperse agent; some nanoparticles have a core size of 24 nm (30%) and the majority have 5 nm cores⁶. The in vitro detection limit using ferucarbotran has been estimated to be approximately 1000 embryonic stem cells (27 pg/cell)⁷. SPIONs designed specifically for MPI are being investigated and show improved performance; in vivo, Wang et al. (2020) demonstrated detection of 2500 bone mesenchymal stem cells with cubic nanoparticles (29 pg/cell)⁸. Beyond this, detection limits for SPION-labeled cells have not been carefully studied.

Sensitivity for ¹⁹F MRI

For ¹⁹F MRI, cells can be labeled with perfluorocarbon agents such as perfluoropolyether (PFPE) nanoemulsions. Since there is little endogenous ¹⁹F in biological tissues, these cells can be visualized with high specificity. The signal intensity of these images is directly linear to the number of ¹⁹F atoms and cell number. The sensitivity of ¹⁹F MRI cell tracking is impacted by ¹⁹F cellular loading. PFPE are formulated into nanoemulsions for safe and effective labeling of cells; clinical-grade PFPE agents are available (CS-1000, CelSense Inc.) and have been used in humans⁹. The amount of ¹⁹F uptake is different for various cell types due to differences in cell size and endocytic ability¹⁰.

MRI hardware and imaging parameters also play a major role in determining ¹⁹F sensitivity. Higher magnetic field strengths improve detectability of cells, additionally, the use of specialized coils is integral. For example, our group has previously demonstrated comparable signal detection at 9.4 T using a birdcage coil compared to 3 T using a surface coil¹¹.

¹⁹F cellular detection limits using various field strengths, hardware, and sequences have been reviewed by Srinivas et al.¹⁰. Notably, there were no reported studies at field strengths ≤ 3 T. The translation of cellular MRI techniques to the clinic will require the use of human MRI systems at clinical field strengths. Our group has demonstrated a cellular detection limit of 25,000 PFPE-labeled macrophages in vitro at 3 T¹¹. In the first human clinical trial at 3 T, an in vivo cellular detection limit between 1 and 10 million dendritic cells was demonstrated⁹. The sensitivity of ¹⁹F MRI for detection of MSC and breast cancer cells at 3 T has not been evaluated.

MPI and ¹⁹F MRI have similar characteristics for cell tracking (positive contrast and quantitation), and MPI is expected to be more sensitive for cell tracking, however, this has not been carefully compared. The objective of this study is to assess the in vitro and in vivo cellular sensitivity of MPI and ¹⁹F MRI for MSC and breast cancer cells.

Methods

Cell culture

4T1 murine breast cancer cells (Dr. Fred Miller, Wayne State University, MI, USA) were maintained in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco, Thermo Fisher Scientific, MA, USA) with 10% fetal bovine serum (FBS) and antimycotic/antibiotic. MSCs derived from the bone marrow of C57BL/6 mice (MUBMX- 01101 [BE], Cedarlane, Burlington, Ontario, Canada) were cultured in low-glucose DMEM (Thermo Fisher Scientific) with 10% FBS. Cells were maintained at 37 °C and 5% CO₂ and passaged every 2–3 days for 10 days.

Cell labeling

2 × 10⁶ 4T1 cells or MSCs were seeded for labeling in T75 cm² flasks. After 24 h, 2.5 mg/mL PFPE nanoemulsion (Cell Sense, Celsense Inc., Pittsburgh, PA, USA) was added to 10 mL complete media and left to co-incubate overnight^12,13. Alternatively, cells were labeled with 55 μg Fe/mL ferucarbotran (Vivotrax, Magnetic Insight Inc., Alameda, CA, USA) using transfection agents in a protocol described by Thu et al.¹⁴ Briefly, 60 μL protamine sulfate (stock 10 mg/mL) was added to 2.5 mL of serum-free medium, and in a second tube, 20 μL heparin (stock 1000 U/mL) and 90 μL ferucarbotran (stock 5.5 mg/mL) was added to 2.5 mL serum-free medium. These two tubes were individually vortexed, then combined. After adhered 4T1 cells or MSC were washed in PBS, this labeling mix was added to the cells. 4 h later, 5 mL of complete media was added to cells and left to co-incubate overnight.

Evaluation of cell labeling

A cytospin of 100 × 10³ cells was prepared for all labeled cells which were fixed in 3:1 methanol:acetic acid. Iron-labeled cells were stained with Perl’s Prussian blue (PPB) to identify iron in cells¹⁵. ¹⁹F-labeled cells were labeled with nuclear fast red to assess for PFPE nanodroplets¹¹. Microscopy of these slides was conducted using the EVOS imaging system (M7000, Thermo Fischer Scientific). The mean intracellular ¹⁹F content in MSC was measured by NMR using a Varian Inova 400 spectrometer (Varian Inc, Paulo Alto, USA) and methods previously described¹¹.

Preparation of cell pellets for imaging

PFPE-labeled cells or ferucarbotran-labeled cells were washed 3 times in phosphate-buffered saline (PBS) and collected in triplicate samples prepared in a dilution series: 1024, 512, 256, 128, 64, 32, 16, 8, 4, 2, 1, 0.5 (× 10³) cells. Ferucarbotran-labeled cells were pelleted in 50 μL PBS for MPI. ¹⁹F-labeled cells were pelleted by centrifugation and covered in 50 μL agarose. This process was repeated two more times to produce a total of 9 replicates for each cell number and cell type (4T1 or MSC), with each cell labeling agent (PFPE or ferucarbotran).

MPI of ferucarbotran-labeled cells

All MPI was conducted on a MOMENTUM™ system (Magnetic Insight Inc.) (Fig. 1A). A small well was secured to the imaging bed where a PCR tube containing the cell pellet could be placed. This allowed for imaging to occur in the identical location and for the user to easily transfer cell samples. Each cell pellet was imaged individually with the following parameters: field of view (FOV) = 6 cm × 6 cm, gradient strength = 3.0 T/m, dual-channel acquisition (X and Z), excitation amplitude = 22 mT (X-channel) and 26 mT (Z-channel), imaging time = 1.5 min. Cell pellets were imaged in descending order of cell number. Once the cell pellet was undetected, 2D imaging with 8 averages (imaging time = 11.8 min) and 3D imaging was conducted for all 9 replicates, in attempt to improve MPI sensitivity. 3D images combine 35 projections in a FOV = 6 × 6 × 6 cm (imaging time = 22.8 min). If cells were detected in these longer scans, fewer cells were imaged until undetected in 2D (8 averages) and 3D.

Figure 1

(A) Image of a preclinical Momentum MPI and sample holder (white arrow), where individual ferucarbotran-labeled cell samples were placed for imaging. (B) Photos of a clinical 3 Tesla MRI with the dual-tuned (¹H/¹⁹F) surface coil, used for ¹⁹F imaging of 6 cell samples. (C) A strong linear relationship exists between MPI signal and iron mass (R² = 0.9997). This was determined by imaging multiple samples of ferucarbotran (0.34–33 μg) and measuring the MPI signal from each sample. The linear equation is subsequently used to convert a measure of MPI signal from ferucarbotran-labeled cells to a measure iron content (μg). (D) ¹H and ¹⁹F images of two reference phantoms, containing known amount of ¹⁹F content (3.33 × 10^{16 19}F spins/μL), are used for quantification of ¹⁹F in cell samples.

¹⁹F MRI of perfluorocarbon-labeled cells

¹⁹F images were acquired on a clinical 3 T MRI (Discovery MR750, General Electric) using a 4.31 × 4.31 cm diameter dual tuned (¹H/¹⁹F) surface coil (Clinical MR Solutions, Wisconsin) (Fig. 1B). Six samples of 32, 64, 128, 256, 512, 1024 (× 10³) cells (from the same replicate) were imaged at a time using a 3D balanced steady state free precession (bSSFP) pulse sequence. ¹⁹F imaging parameters were: FOV = 40 × 20 mm, matrix = 40 × 20, slice thickness = 1 mm (1 × 1 × 1 mm³ resolution), repetition time/echo time = 5.6 ms/2.8 ms, bandwidth = ± 10 kHz, and flip angle = 72°. Flip angle was optimized through a series of ¹⁹F images (Fig. S1) and calculated using reported T1 and T2 times at 3 T^9,16. Cell pellets were imaged with 115 excitations (imaging time of 9.5 min, or 1.5 min/pellet) and 345 excitations (imaging time of 28.3 min, or 4.5 min/pellet).

In vivo detection of MSC using MPI and ¹⁹F MRI

Seven NOD/SCID/ILIIrg − / − (NSG) mice were obtained and cared for in accordance with the standards of the Canadian Council on Animal Care. The study was approved by the Animal Use Subcommittee of Western University’s Council on Animal Care. All methods were carried out in accordance with relevant guidelines and regulations. This study was carried out in compliance with the ARRIVE guidelines.

In preparation for MPI, 4 mice were fasted for 12 h, by removing food pellets from their cages, and a piece of ex-lax® (regular strength stimulant laxative) was added to the cage. Additionally, cotton bedding, which has been shown to produce MPI signal (unpublished results) was replaced with corn bedding. The goal of these measures was to minimize background signal associated with the iron in the mouse digestive system.

To investigate differences in sensitivity in vivo, 1 × 10⁵ ferucarbotran-labeled MSC were injected to NSG mice by subcutaneous, intraperitoneal, or intravenous injection. A fourth mouse received 2 × 10⁶ ferucarbotran-labeled MSC by subcutaneous injection. Each mouse was imaged with MPI with a FOV = 12 cm × 6 cm × 6 cm in 2D (2.2 min). All other imaging parameters are described above. MPI was conducted immediately before injections, to measure background signal, and immediately after injections. MPI signal for each mouse was calculated as the difference between pre-injection signal (background) and post-injection signal. MPI signal from cells in vivo was compared to MPI signal from cell pellets.

Similarly, 2 × 10⁶ PFPE-labeled MSC were injected to NSG mice by subcutaneous or intraperitoneal injection. A third mouse received fewer PFPE-labeled cells (1 × 10⁵) by subcutaneous administration. Following cell administration, ¹⁹F MRI was conducted for each mouse by placing the surface coil directly above the injection site to maximize sensitivity (shown in Fig. 7E); the mouse receiving cells by subcutaneous was prone for imaging and the mouse receiving cells by intraperitoneal injection was supine. Imaging parameters are the same as listed above, however with a FOV = 60 × 30 mm with matrix size = 60 × 30 (1 × 1 × 1 mm³ resolution), and 200 excitations (imaging time = 35.5 min). In vivo ¹⁹F signal was quantified (described below) and compared to signal detected from cell pellets, which were included alongside the mouse.

Image analysis

For MPI signal calibration, an additional 8 samples of ferucarbotran (5.5 mg/mL) were imaged with identical parameters in a dilution series in PBS: 33, 22, 11, 5.5, 2.25, 1.38, 0.68, 0.34 μg ferucarbotran. A linear relationship was found between iron mass and MPI signal (Fig. 1C) and the equation of the line was used to calculate associated iron content from each cell sample.

2D MPI of the empty sample holder was conducted at the beginning (0 h), middle (3 h), and end (6 h) of six imaging sessions. The standard deviation of background noise (SD_noise) was measured in these 18 images. To quantify signal from ferucarbotran-labeled cells in pellets and in vivo, a threshold of 5 times the average background SD_noise was used to mask lower amplitude signal and yield a reliable measurement of MPI signal. This imaging criteria is based on MPI signal with SNR > 5 (Rose Criterion)¹⁷. Total MPI signal was calculated as mean MPI signal * volume of ROI (mm² or mm³).

Delineation of ¹⁹F signal was conducted using a similar method as MPI (5*SD_noise). Background SD_noise of ¹⁹F signal for each 3D image was measured by drawing an ROI in background noise of one image slice. Due to Rician distribution observed in background signal noise, ¹⁹F signal between 5*SD_noise and 8*SD_noise was corrected using a factor of 0.655, as described by Bouchlaka et al. (2016)¹⁸ (Fig. S2). Total ¹⁹F signal was calculated as mean ¹⁹F signal * volume of ROI. Two reference phantoms containing 3.33 × 10^{16 19}F spins/μL were included in the imaging field of view for calibration (Fig. 1D). ¹⁹F content from cell pellets was calculated as:

$$19F\;spins\;in\;ROI = \left( {signal\;in\;ROI} \right)*\left( {3.33 \times 10^{16} \frac{19F\;spins}{{{\mu L}}}} \right)*\left( {\frac{{volume\;of\;phantoms\;in\;mm^{3} }}{average\;signal\;in\;phantoms}} \right)$$

We defined MPI and ¹⁹F MRI in vitro cell detection limits as the minimum number of MSC and 4T1 cells detected with SNR > 5. Thus, cells with signal below the 5*SD_noise criteria were considered undetected. The amount of ferucarbotran or PFPE associated with the lowest cell number was calculated as iron mass per cell * number of cells or ¹⁹F atoms per cell * number of cells, respectively.

Statistical analysis

All statistical analysis were performed using GraphPad Prism version 9. Linear regression was performed for MPI calibration (known iron mass vs. measured MPI signal) to determine the calibration equation. This line is forced through the origin, under the assumption that background MPI signal, without a sample of iron, has an average of 0. Pearson’s correlation was conducted for MPI (number of cells vs. measured MPI signal) and ¹⁹F MRI (number of cells vs. measured ¹⁹F signal). Analysis of co-variance (ANCOVA) was used to evaluate whether the MPI sensitivity (slope) was significantly different for ferucarbotran-labeled 4T1 cells and MSC (number of cells vs. measured MPI signal). Analysis of variance (ANOVA) was used to analyse differences between MPI signal measured from each cell number and again for ¹⁹F signal measured from each cell number. A p-value of 0.05 was used to determine statistical significance, unless otherwise indicated.

Results

MPI cellular sensitivity

The detection of ferucarbotran-labeled cells using 2D MPI (imaging time = 1.5 min) is shown in Fig. 2A (4T1 cells) and Fig. 3A (MSC). MPI signal measured from samples of 8000–1,024,000 4T1 cells and 4000–1,024,000 MSC had SNR > 5. The iron mass significantly increased with cell number for both cell types (Figs. 2A, 3A). Importantly, the standard deviation of background noise, measured three times over six imaging sessions, showed no significant differences (Fig. S3).

Figure 2

MPI detection of ferucarbotran-labeled 4T1 cells. (A) 2D MPI of individual 4T1 cell pellets containing the indicated cell numbers (M = 10⁶, k = 10³). As few as 8000 cells (0.074 μg ferucarbotran) could be detected with SNR > 5 in 1.5 min. 2D MPI signal (and the associated iron content) significantly increases with the number of 4T1 cells (*P < .05, **P < .01, ***P < .001, ****P < .0001). (B) With 8 signal averages, the same result was found. (C) In 3D, the detection of 4000 cells (0.037 μg ferucarbotran) with SNR > 5 was possible in 23 min. Images with SNR < 5 have white boxes to indicate the placement of the cell sample.

Figure 3

MPI detection of ferucarbotran-labeled MSC. (A) 2D MPI of individual MSC pellets containing the indicated cell numbers (M = 10⁶, k = 10³). As few as 4000 MSC (0.076 μg ferucarbotran) could be detected with SNR > 5 in 1.5 min. 2D MPI signal (and the associated iron content) significantly increases with the number of MSC (*P < .05, **P < .01, ***P < .001, ****P < .0001). (B) With 8 signal averages, the same result was found. (C) In 3D, the detection of 2000 cells (0.038 μg ferucarbotran, with SNR > 5) and 1000 cells (0.019 μg ferucarbotran, with SNR > 3) was possible in 23 min. Images with SNR < 5 have white boxes to indicate the placement of the cell sample.

MPI signal (and the associated iron content) was strongly correlated with cell number for ferucarbotran-labeled 4T1 cells and MSC (Fig. 5A). The slope of the line was significantly steeper (factor of 2.07) for MSC compared to 4T1 cells (P < 0.0001). Enhanced ferucarbotran labeling was measured in MSC (19.09 ± 2.50 pg Fe/cell) compared to 4T1 cells (9.22 ± 1.42 pg Fe/cell), which can be visualized in PPB-stained cells (Fig. 5B).

With 2D MPI, the lowest cell numbers detected with SNR > 5 were 8000 4T1 cells (6/9 replicates, corresponding to 74 ng iron) and 4000 MSC (8/9 replicates, corresponding to 76 ng iron). All other replicates of 8000 4T1 cells and 4000 MSC had SNR > 3. For both cell types, 2D imaging with 8 averages did not improve cell detection (Figs. 2B, 3B). However, additional averaging does reduce the standard deviation of background signal from 0.235 (1 average) to 0.183 (8 averages) (P < 0.0001).

In 3D images (22.8 min), detection of 4000 4T1 cells (37 ng iron) was enabled (3 of 9 replicates SNR > 5, and 8 of 9 had SNR > 3) (Fig. 2C). Improvements to detection of MSC was also seen in 3D; 2000 MSC (38 ng iron) were detected (6 of 9 replicates SNR > 5, and 9 of 9 had SNR > 3) (Fig. 3C). In 4 of 9 replicates, the detection of 1000 MSC (19 ng iron) was possible in 3D with a threshold of SNR > 3.

¹⁹F MRI Cellular Sensitivity

MSC were labeled with 3.52 ± 1.55 × 10^{11 19}F/cell (NMR) and the nanodroplets associated with the ¹⁹F agent were identified with microscopy (Fig. 5D). The average number of ¹⁹F spins determined by NMR was not significantly different from that measured by MRI in both short (P = 0.3648) and long scans (P = 0.8541). 4T1 cells did not label sufficiently with PFPE, as ¹⁹F content in 1 × 10⁶ cells was undetected by NMR. Thus, ¹⁹F imaging experiments with PFPE-labeled 4T1 cells did not continue.

In ¹⁹F images (with imaging time = 1.5 min/pellet), the range of 256–1024 × 10³ PFPE-labeled MSC were detected (Fig. 4A) and ¹⁹F signal was strongly correlated with cell number (R² = 0.9983) (Fig. 5C). The average number of cells detected per voxel from these scans was 1.30 ± 0.51 × 10⁵ cells/mm³. The lowest cell number detected from these ¹⁹F scans was 256 × 10³ MSC (9.01 × 10^{16 19}F atoms, 30.21 mM), which was detected in 4 of 9 replicates with SNR > 5. ¹⁹F signal measured from 1.024 × 10⁶ MSC was significantly higher than signal measured from 512 × 10³ cells (P < 0.01), which was higher than signal measured from 256 × 10³ cells (P < 0.05) (Fig. 4C).

Figure 4

¹⁹F MRI detection of PFPE-labeled MSC. (A) ¹⁹F images of six samples with various cell numbers (M = 10⁶, k = 10³) imaged 1.5 min/pellet. As few as 256 × 10³ cells (9.01 × 10^{16 19}F atoms) could be detected with SNR > 5. (B) With longer imaging time (4.5 min/pellet), the detection of 128 × 10³ cells (4.51 × 10^{16 19}F atoms) was possible with SNR > 5. (C) Quantification revealed significant differences in ¹⁹F signal between different numbers of MSC (*P < .05, **P < .01). Significantly more ¹⁹F signal was detected from 1.024 × 10⁶ cell samples with longer imaging times (P < .01).

Figure 5

(A) The number of ferucarbotran-labeled MSC and 4T1 cells is strongly correlated with iron content measured by MPI (R² > 0.998). The slope of the line for MSC is higher than 4T1 (P < .0001), indicating higher sensitivity and enhanced uptake of iron in MSC compared to 4T1 cells as shown with PPB stain (B). (C) A strong linear correlation exists between the number of PFPE-labeled MSC and detected ¹⁹F signal (R² = 0.9983). (D) PFPE labeling identified as nanodroplets in microscopy (black arrows).

With longer imaging times (4.5 min/pellet), ¹⁹F sensitivity was improved (Fig. 4B) and the SD_noise was significantly reduced compared to 1.5 min/pellet scans (P < 0.0001). The average number of cells detected per voxel from these scans was 8.55 ± 2.97 × 10⁴ cells/mm³. As few as 128 × 10³ MSC (4.51 × 10^{16 19}F atoms, 19.01 mM) were detected with SNR > 5 in 3 of 9 replicates. Additionally, ¹⁹F signal with SNR > 5 was detected in 7 of 9 replicates of 256 × 10³ MSC, which corresponds to 9.01 × 10^{16 19}F atoms (28.01 mM). Significant differences in ¹⁹F signal were measured from 256, 512, and 1024 (× 10³) MSC (Fig. 4C). Furthermore, significantly more ¹⁹F signal was measured from 1.024 × 10⁶ cells when using longer imaging times (4.5 min/pellet) compared to 1.5 min/pellet (P < 0.01).

In vivo sensitivity of MPI and ¹⁹F MRI

A comparison between MPI signal from a cell pellet and cells in vivo was conducted with different injection routes (Fig. 6A). These MSC were labeled with 28.9 ± 3.4 pg iron/cell. Compared to a pellet of 2 × 10⁶ ferucarbotran-labeled MSC, MPI signal was only reduced by 5% with subcutaneous injection of these cells (Fig. 6 B,C). Quantification revealed the iron mass measured from the cell pellet (52.98 μg) was similar to what was measured after subcutaneous injection (50.21 μg). However, for 1 × 10⁵ MSC, in vivo MPI showed a reduction in MPI signal measured from MSC injected subcutaneously (49%), intraperitoneal (53%), and intravenous (15%), compared to signal from a pellet of 1 × 10⁵ MSC (Fig. 6 D,E). For 1 × 10⁵ MSC, the measured iron content was 3.13 μg in the cell pellet, compared to 1.52 μg after subcutaneous injection, 1.66 μg after intraperitoneal injection, and 0.48 μg after intravenous injection. Therefore, the iron content measured from 1 × 10⁵ cells in vivo was reduced compared to cells in the pellet, despite being the same number of cells. After mouse fasting, the background in vivo MPI signal from the mouse digestive system was 25.8 ± 10.0 arbitrary units (A.U.) (shown Fig. 6F). This background signal was accounted for in each mouse by signal subtraction, prior to calculation of MPI signal and iron mass measured from cells.

Figure 6

(A) Cartoon showing subcutaneous (SQ), intraperitoneal (IP), and intravenous (IV) injection in mice. (B) Comparison of MPI signal from 2D scans of 2 × 10⁶ ferucarbotran-labeled MSC in a cell pellet to subcutaneous injection. (C) Measured MPI signal from 2 × 10⁶ MSC was similar in vivo and ex vivo. (D) Detection in 2D of 1 × 10⁵ ferucarbotran-labeled MSC as a cell pellet and after subcutaneous, intraperitoneal, and intravenous injection. (E) Measured MPI signal from 1 × 10⁵ cells is reduced in vivo compared to signal in the pellet. (F) Some background MPI signal exists owing to iron in mouse digestion. This signal was minimized with 12-h fasting and accounted for by subtracting pre-injection signal from post-injection signal for each mouse.

The detection of 2 × 10^{6 19}F-labeled MSC in vivo was compared to MSC in a pellet (Fig. 7). Reduced ¹⁹F signal (72%) was detected from MSC following subcutaneous injection (Fig. 7A, D). After intraperitoneal injection, the same number of cells were dispersed and appeared as lower intensity ¹⁹F signal (Fig. 7B), however higher ¹⁹F signal was measured from these cells compared to the pellet by 6.65 times (Fig. 7D). 1 × 10⁵ MSC administered subcutaneously were undetected as this cell number is below the detection limit (Fig. 7C).

Figure 7

Detection of 2 × 10⁶ PFPE-labeled MSC in vivo. ¹⁹F signal is detected at the injection site following (A) subcutaneous injection (white arrows) and (B) intraperitoneal injection (white ovals). A cell pellet of the same cell number was imaged alongside the mouse for comparison (shown in A). (C) After subcutaneous administration, 1 × 10⁵ cells were undetected, as this cell number is below the detection threshold for MSC. Images are sagittal (left) and axial (right). M = million, k = thousand. (D) ¹⁹F signal measured from 2 × 10⁶ cells injected subcutaneous (SQ) was reduced compared to signal measured from the cell pellet, however, elevated ¹⁹F signal was measured from cells following intraperitoneal (IP) injection. (E) The dual-tuned surface coil is approximately the same size as a mouse and is placed directly over the injection site for imaging.

Discussion

In this study, we began with an evaluation of in vitro sensitivity for MPI and ¹⁹F MRI of cells using ferucarbotran and PFPE nanoemulsions as labeling agents (respectively). Overall, fewer MSC were detected using MPI (4000) compared to ¹⁹F MRI (256,000) using the same imaging time (1.5 min per cell pellet). Compared to ferucarbotran-labeled MSC, more 4T1 cells were required for MPI detection (8000) as a result of lower cell uptake of ferucarbotran. These limits were defined with imaging criteria SNR > 5 and tested with 9 replicates to provide confidence.

These measurements of lowest cell number detected with MPI and ¹⁹F MRI are reasonable based on previous reports. As described earlier, Zheng et al.⁷ achieved detection of approximately 1000 ferucarbotran-labeled human embryonic stem cells (27 ng $n$/cell, 27 ng) with SNR > 5. Our MSC detection limit was higher (4000 cells or 76 ${\text{n}}$g) and this is resulting from differences in MPI systems and acquisition. The amount of iron we detected (76 ng) is consistent with findings by Liu et al.¹⁹, where 64 ng ferucarbotran could be detected with mean SNR of 3.6, using another MOMENTUM MPI system and the same 2D imaging parameters. With longer imaging times in 3D, we demonstrated a minimum of 1000 ferucarbotran-labeled MSC (19 ${\text{n}}$g) could be visualized with SNR > 3. To the best of our knowledge, this is the lowest report of ferucarbotran detected from labeled cells. This study was the first to measure MPI detection limits for iron-loaded cancer cells (in 3D, as few as 4000 4T1 cells were detected, or 37 ${\text{n}}$g iron).

Previous investigation into ¹⁹F detection limits at 3 T were conducted in dendritic cells (3.7 × 10^{12 19}F/cell)⁹ and macrophages (7.93 × 10^{11 19}F/cell)¹¹, which have greater uptake of PFPE than what we measured in MSC (3.52 × 10^{11 19}F/cell). Ahrens et al. established an in vitro ¹⁹F MRI detection limit of 1.0 × 10⁵ dendritic cells/voxel on a 3 T clinical scanner using a 2.5*SD_noise threshold⁹. Similarly, we achieved detection of 1.30 × 10⁵ MSC/voxel (1.5 min/pellet) and 8.55 × 10⁴ MSC/voxel in longer scans (4.5 min/pellet) using a 5*SD_noise threshold. In our previous work at 3 T, as few as 25,000 murine macrophages were detected (2.27 × 10⁴ cells/voxel) with longer imaging time¹¹.

Cell labeling is fundamental in determining cellular sensitivity for both MPI and ¹⁹F MRI. We previously discussed factors that impact labeling efficiency with SPIONs²⁰. In this study, we observed enhanced MPI detectability of MSCs compared to breast cancer cells, owing to increased endocytosis of SPION. Likewise, 4T1 cells did not label with PFPE sufficiently for ¹⁹F MRI detection. Increased uptake of SPION/PFPE is expected in phagocytic cells and cells which have a larger cytoplasmic volume²¹, such as macrophages and dendritic cells. Although MSC is considered a non-phagocytic cell type²², enhanced labeling compared to breast cancer cells has been observed with SPIONs¹⁵ and PFPE¹⁰. The cell labeling protocols used in this study use commonly used concentrations of ferucarbotran and PFPE for cell labeling and have been well established. We, and other groups, have not observed compromises in cell viability or function at these labeling concentrations, even in delicate cell types^12,13,23–²⁷. In this study, we use transfection agents to facilitate uptake of ferucarbotran. It is expected that SPION surface modifications will be improve labeling efficiency without transfection agents^8,28. Further improvements to PFPE nanoemulsions will enhance ¹⁹F cellular sensitivity, such as incorporation of paramagnetic agents^29,30, or the addition of surface modifications to enhance the uptake of PFPE nanoemulsions³¹.

We recognize it is impossible to directly compare cell detection with MPI and ¹⁹F MRI due to their inherent differences, including structural configurations and imaging parameters. For this study, we attempted to optimize unique aspects of each modality in favor of sensitivity. MPI of cells was conducted using weak gradients (3 T/m) to increase the size of the FFR and enhance sensitivity. For FFL (line) MPI, the use of weak gradients (3 T/m) compared to stronger gradients (e.g. 6 T/m) expands the volume of FFR. This leads to enhancement in sensitivity, at the cost of reduced resolution. It is also expected that signal averaging would improve sensitivity, however this has not been studied for MPI. In 2D, a significant reduction in background noise was measured with 8 averages compared to 1 average, however, this did not improve cell detection with MPI, as we defined it. 3D imaging using 35 projections did offer improvement in sensitivity for both ferucarbotran-labeled MSC (detection of 2000 cells) and breast cancer cells (4000 cells). Lastly, optimization of excitation amplitudes may lead to improved cell sensitivity; in this study we used 22 mT (X-channel) and 26 mT (Z-channel) by default.

¹⁹F MRI of PFPE-labeled cells was conducted at a clinical field strength (3 T). The implementation of the surface coil and optimized 3D bSSFP sequence is crucial to enable ¹⁹F cell tracking at 3 T. The theoretical optimal flip angle for ¹⁹F at 3 T is 72° and our investigation showed highest SNR was produced for flip angles between 60—80°. However, transmit/receive surface coils provide non-uniform sensitivity, due to spatial variations in applied energy and flip angle ^32,33. For this reason, cell pellets were imaged directly in the center of the coil to maximize sensitivity. Likewise, the surface coil was placed directly above the region of interest for in vivo ¹⁹F imaging. High signal averaging was used for ¹⁹F image acquisition (115) to improve SNR. Longer imaging times with 345 signal averages enabled detection of 3 additional pellets of 256 × 10³ cells and 3 of 9 pellets of 128 × 10³ PFPE-labeled MSC.

After assessing in vitro cell detection limits of MPI and ¹⁹F MRI, a preliminary assessment of in vivo detection factors was conducted. It has previously been shown that there is no attenuation of MPI signal from biological tissue^34,35. In agreement, following subcutaneous injection of 2 × 10⁶ MSC, we measured only a small reduction in cell detection with MPI (5%). However, for a lower cell number (1 × 10⁵), there was reduction of signal measured in vivo compared to the pellet (by 47—85%, depending on injection route). Here we recognize that the dispersion of cells from the injection site reduces the cell density per voxel, leading some cells to fall below the intravoxel detection limit. MPI detection of MSC was most reduced after intravenous injection (85% reduction). It is expected that cells administered intravenously would be most disperse as they circulate through the venous circulation before accumulating in the lung capillaries (shown in Fig. 6A). Similarly, previous work by Wang et al. (2020)⁸ showed that 1 × 10⁵ ferucarbotran-labeled stem cells could be detected with MPI in vivo following subcutaneous injection but not following intravenous injection. In our study, we could achieve detection of 1 × 10⁵ cells after intravenous injection, which can be attributed to the choice of gradient strength (3.0 T/m vs. 5.7 T/m).

Likewise for ¹⁹F MRI, dispersion of PFPE-labeled cells (1 × 10⁶) in patients was previously reported to limit the detectability of these cells following administration⁹. In our study, ¹⁹F signal detected from a cell pellet of 2 × 10⁶ PFPE-labeled MSC was higher than ¹⁹F signal from the same number of cells injected subcutaneously. Conversely, ¹⁹F signal measured from cells injected to the intraperitoneal space was overestimated. This result could be explained by the large quantification region, as ¹⁹F density per imaging voxel was 3.79 × 10^{16 19}F/mm³ for cells in a pellet, compared to 2.03 × 10^{16 19}F/mm³ for cells in vivo.

For both MPI and MRI, there are other important considerations which reduce detectability of cells in vivo. For MPI, this includes increased background signal associated with mouse digestion³⁶. While we accounted for background signal in the mouse using signal subtraction, this technique is not permissible for longitudinal cell tracking studies. Background signal is variable across mice and day-to-day. In our experience, this can be reduced by mouse fasting, however, the amount of signal is unpredictable. Ultimately this background signal may obscure detection of cells, especially in low cell numbers³⁷, as it is challenging (at this time, impossible) to distinguish the signal associated with cells from background. Second, there is some evidence that Brownian relaxation of SPION in different tissue environments may be altered, leading to reduced MPI sensitivity³⁸–⁴⁰. Brownian motion refers to the physical rotation of SPIONs; this motion is reduced in tissues with increased stiffness (e.g. muscle) which leads to increased Brownian relaxation times (thus, lower sensitivity and resolution). Ongoing work aims to determine whether this plays a role in the detection of SPION-labeled cells. Another consideration affecting ¹⁹F detection of PFPE-cells in vivo is the coil filling factor. The volume of a mouse is much larger than the volume of a cell pellet, thus SNR for detection of cells in vivo is expected to be reduced due to increased image noise. Lastly, mouse breathing motion can lead to blurring of signal which will reduce the maximum signal intensity associated with cells and could potentially render cells undetected by MRI and MPI.

Conclusion

MPI has the potential to be more sensitive than ¹⁹F MRI for cell tracking. In this study, commonly used cell labeling agents, ferucarbotran and perfluoropolyether nanoemulsions, were used to label MSC and 4T1 cells to assess cellular sensitivity of MPI and ¹⁹F MRI, respectively. Fewer MSC (4000 cells, 76 ng iron) were detected with MPI than ¹⁹F MRI (256,000 cells, 9.01 × 10^{16 19}F atoms), using the same scan time. Furthermore, reduced ferucarbotran labeling was observed in 4T1 breast cancer cells compared to MSC, leading to a detection limit of 8000 breast cancer cells (74 ng iron). With longer imaging times, as few as 2000 MSC (38 ng ferucarbotran) and 4000 breast cancer cells (37 ng ferucarbotran) were detected with MPI and 128,000 MSC (4.51 × 10^{16 19}F atoms) were detected with ¹⁹F MRI with SNR > 5. Determination of these detection thresholds in vitro is useful to anticipate the minimum number of cells that are required for detection in vivo. However, we demonstrated that there are several factors in vivo which led to reduced detectability of cells, particularly the effect of cell dispersion which reduces cell density per imaging voxel. There is no doubt that cellular sensitivity for these modalities will continue to improve with further developments. It is essential to understand and improve cellular sensitivity to advance imaging of cellular therapeutics.