Whole-body in vivo MPI cytometry reveals injection route-, dose-, cell size–, and disease-dependent differences in organ distribution

Abstract

Robust quantification of the whole-body biodistribution of injected therapeutic cells remains a major challenge. We performed in vivo cytometry using magnetic particle imaging (MPI) to track magnetically labeled cells on a timescale of minutes with high sensitivity, zero background signal, and simple linear quantification. Human mesenchymal stem cells (hMSCs; ~25 micrometers) and human neural precursor cells (~10 micrometers) were labeled with superparamagnetic iron oxide nanoparticles and mapped with MPI after intravenous or intra-arterial injection in normal mice. The tissue site and numbers of cell accumulation and retention were dependent on cell size, dose, injection frequency, and injection route, with the lung and liver serving as the major entrapment organs. In mice with experimental autoimmune encephalomyelitis, but not in normal mice, hMSCs also homed to the spleen. Integrating MPI cytometry with cell therapy may aid in further optimization of the route, dose, and frequency of cell administration.

Whole-body in vivo cytometry using MPI is able to dynamically track and quantify tissue cell distribution.

INTRODUCTION

Developing noninvasive in vivo cytometry methods to longitudinally track and quantify therapeutic stem cells and immune cells remains an active area of research to successfully translate cell therapies into clinical practice (1). It has been shown that cell injections can miss their target in up to 50% of patients even when injected under ultrasound imaging–guided injections (2), emphasizing the need for more effective methods to monitor and improve cell delivery (3). Traditional imaging modalities for in vivo cytometry or quantitative cell tracking often fall short in providing the necessary sensitivity, specificity, and quantification required for precise whole-body cell tracking. ¹H magnetic resonance imaging (MRI) cell tracking (2, 4, 5) lacks the specificity and accuracy needed to quantify magnetically labeled cells with reliable whole-body biodistribution after systemic delivery. Nuclear imaging, either positron emission tomography (PET) or single-photon emission computed tomography (SPECT), is a quantitative method and offers high sensitivity for cell tracking but involves ionizing radiation and lacks spatial resolution (6). Optical imaging suffers from limited tissue penetration and quantification difficulties (7). Magnetic particle imaging (MPI) has emerged as a promising modality for in vivo cell tracking (8–11) with a reported in vivo detection sensitivity as low as 250 labeled cells (12). MPI directly detects superparamagnetic iron oxide (SPIO) nanoparticles with high sensitivity and specificity (13, 14). MPI provides a linear relationship between signal intensity and SPIO concentration, enabling accurate quantification of SPIO-labeled cells and tracers (8, 10, 15). ¹⁹F MRI has been explored as an alternative quantitative cell tracking approach for in vivo cytometry (16), but it suffers from lower sensitivity and temporal resolution compared to MPI (17). The main limitations of MPI include its relatively low spatial resolution and its lack of anatomical context, necessitating its combination with imaging modalities such as MRI or computed tomography (CT) for accurate anatomical localization. Recent progress in the development of human-scale MPI scanners demonstrates that the technology is scalable and positioned for clinical translation (18–21). The use of nonionizing magnetic fields in MPI (22), combined with the biocompatibility and minimal functional impact of the magnetic labeling of cells (23), supports the future use of clinical MPI for real-time cell tracking.

Here, we introduce the concept of in vivo MPI cytometry, an imaging approach that enables dynamic whole-body tracking and quantification of SPIO-labeled therapeutic cells in living subjects on a timescale of minutes. A schematic overview of the MPI cytometry workflow is shown in Fig. 1 to illustrate the experimental sequence and integration of various methodological components.

Fig. 1. Schematic overview of the MPI cytometry setup and experimental workflow.

IV, intravenous; IA, intra-arterial. Created in BioRender. A. Shakeri-Zadeh (2025), https://biorender.com/jy252vx.

Implementing the labeled cell-fiducial calibration strategy, in which reference tubes containing known numbers of labeled cells from the same batch used for injection are imaged alongside each animal, is central to our MPI cytometry workflow and enables direct conversion of organ MPI signal into cell numbers both in vivo and ex vivo. Most if not all studies as of today have used fiducials containing simple aqueous solutions of suspended SPIO particles, but it has now become clear that these fiducials do not mimic labeled cells (24) and tissues (25). In vivo MPI cytometry was performed on two distinct stem cell types, i.e., human mesenchymal stem cells (hMSCs), which are relatively large cells (~25 μm in diameter or 8000 μm³ for a spherical volume, which the cells have following trypsinization before injection) used before in preclinical MPI (8, 26) and clinical MRI (27) cell tracking studies, and human neural precursor cells (hNPCs), representing smaller cells (~10 μm in diameter or ~500 μm³ in volume) used with MRI cell tracking (28). Selecting these two distinct human stem cell types allows a direct comparison of the impact of cell size and delivery route on whole-body biodistribution. Our results reveal distinct biodistribution profiles for the two cell types, highlighting the suitability of MPI cytometry for quantitative monitoring of cell distribution in vivo. Although MPI has been previously used to track cells in vivo (10, 26, 29), cells were injected intraparenchymally in tissue (no systemic injection) or imaged at much later time points without initial fast dynamic quantitative imaging at a timescale of minutes. To broaden the biological relevance of in vivo MPI cytometry and highlight its utility beyond assessment of the baseline biodistribution, we also applied the method in an experimental autoimmune encephalomyelitis (EAE) neuroinflammatory disease model and were able to identify the spleen as a key site of hMSC accumulation following systemic administration. The detection of SPIO-labeled hMSCs in the spleen, validated by both MPI and histological analysis, supports the concept of systemic immunosuppression (30–32), wherein hMSCs exert antineuroinflammatory effects through remote splenic cell-cell interactions. This observation underscores the potential of MPI to further uncover mechanistic insights into the biodistribution and therapeutic action of systemically delivered stem cells.

RESULTS

Whole-body quantitative in vivo MPI cell tracking

hMSCs were efficiently labeled with SPIO (Fig. 2A). In vivo MPI/CT imaging revealed distinctly different biodistribution profiles for intravenous compared to intra-arterial injection (Fig. 2, B and C). At 30 min post–intravenous injection, cells accumulated within the lungs, with negligible signal in the liver and brain, indicating pulmonary entrapment. In contrast, intra-arterial injection revealed cell localization in the brain, lungs, and liver. After 1 day, intravenously injected cells completely redistributed to the liver, while intra-arterially injected cells persisted in both the brain and liver without lung signal (Fig. 2B). Using a calibration curve for the same batch of labeled hMSCs that were injected, the numbers of cells at 30 min post–intra-arterial injection were calculated to be 99,937 ± 10,392 cells for the lung/liver and 27,500 ± 2240 cells for the brain. These total numbers summing up to ~125,000 cells closely equals the amount of 120,000 injected cells, translating to an iron mass of 57 pg Fe per cell as estimated from signal intensity and known Fe content determined using a ferrozine assay.

Fig. 2. Whole-body in vivo tracking and quantification of hMSCs following intravenous or intra-arterial injection.

(A) Macro- and microscopic visualization of ferucarbotran-labeled hMSCs before injection. (B) In vivo MPI/CT images of labeled hMSCs at 30 min and 1 day after intravenous or intra-arterial injection. S.I., signal intensity. (C) Quantitative analysis of cell accumulation 30 min post–intra-arterial injection. Two fiducials (F1 = 10,000 and F2 = 20,000 labeled cells) were used for calibration (y = 0.13x − 94.92; R² = 0.98). Data in each figure panel represent three independent experiments. B, Lu, Li, and F stand for the brain, lungs, liver, and fiducials, respectively.

Bimodal MPI/MRI cell tracking

To validate the cerebral delivery of ferucarbotran-labeled hMSCs following intra-arterial injection, we performed ex vivo magnetic imaging (Fig. 3A) and histological analysis (Fig. 3B). Coronal MRI scans showed unilateral hypointense regions across multiple brain slices, consistent with labeled cell accumulation. The signal voids were present from midbrain to forebrain, indicating a global cell distribution in the entire ipsilateral hemisphere. MPI scans of the same brains revealed hotspots that overlapped with the MRI hypointensities. Histological Prussian blue (PB) staining further confirmed the presence of SPIO-labeled hMSCs. Anti–human nuclear antigen (HuNa) staining revealed that the iron-loaded cells contained human nuclei, confirming that they were the hMSCs that were injected, corroborating the MRI/MPI findings.

Fig. 3. Bimodal MPI/MRI of hMSCs at 6 hours after intra-arterial injection.

(A) Consecutive coronal ex vivo MRI, MPI, and MRI/MPI overlay images showing MRI hypointensities (white arrows) and MPI hotspots from midbrain to forebrain, corresponding to labeled hMSCs. (B) Corresponding coronal brain sections stained for iron (top row) and anti-HuNa (bottom row), demonstrating that the remains localized in hMSCs.

Longitudinal MPI cell tracking

To investigate the long-term biodistribution and persistence of intra-arterially injected cells, we performed serial in vivo and ex vivo MPI scans at 6 hours (day 0), 2 days, 7 days, and 30 days postinjection of 120,000 SPIO-labeled hMSCs (Fig. 4A). Using two fiducials containing labeled hMSCs from the same batch that was injected, MPI signal quantification revealed 22,294 ± 1920 cells in the brain and 114,940 ± 13,125 cells in the lungs/liver on day 0 (Fig. 4B). By day 2, the brain signal became undetectable in vivo, indicating that most cells redistributed from the brain to other organs. The lungs and liver retained 117,385 ± 14,450 cells at day 2, reflecting the early loss of brain-localized cells and sustained hepatic/pulmonary retention. The cell number in the liver declined to 58,835 ± 5210 at day 7 and 24,280 ± 2225 at day 30. Ex vivo MPI confirmed these observations with no detectable cells in the heart. At day 0, 8200 ± 970 cells were detected in the brain, 5430 ± 436 in the lungs, and 102,500 ± 11,240 in the liver, providing a (lung + liver)/brain ratio of ~13. By day 2, only 2330 ± 190 cells remained in the brain, and no cells were detected in the lungs, while the liver signal increased to 108,800 ± 8425 cells, yielding a (lung + liver)/brain ratio of ~47. At days 7 and 30, the liver retained 67,300 ± 7360 and 22,240 ± 2015 cells, respectively, with no detectable cells in the lungs or brain. These findings suggest a possible redistribution of SPIO-labeled cells from the lungs and brain into the liver between days 0 and 2, followed by gradual clearance.

Fig. 4. Longitudinal MPI of labeled hMSCs after intra-arterial injection.

(A) In vivo and ex vivo MPI at different time points postinjection. Color-coded heatmaps indicate signal intensity. Dotted circles highlight organs with no detectable signal. Shown are representative data from three independent experiments. h, hours. (B) Ex vivo MPI-based quantification of labeled hMSCs. Similar to the in vivo studies, F1 = 10,000 and F2 = 20,000 labeled cells were used as fiducials for ex vivo quantification.

Fast dynamic whole-body in vivo MPI cytometry: Effect of cell size

To compare the in vivo biodistribution of two stem cell types having different sizes, we performed dynamic MPI tracking of 120,000 intra-arterially injected ferucarbotran-labeled hMSCs and 960,000 Synomag-D70–labeled hNPCs, which have spherical diameters of ~25 and 10 μm following trypsinization, respectively (Fig. 5). Each total cell dose was injected using four incremental injections, with equal cell numbers delivered in each injection. Both cell types were effectively labeled as confirmed with PB staining. For hMSCs, serial in vivo MPI performed over 12 min postinjection revealed a prominent signal in both the brain and liver/lung regions, with the signal increasing over time from 0 to 12 min. Using a calibration curve obtained from fiducials containing known numbers of labeled cells from the same batch as was injected, it could be calculated that 22,569 ± 1805 hMSCs localized in the brain, with 112,808 ± 10,600 cells in the combined liver and lung regions, yielding a (lung + liver)/brain ratio of ~5. The labeled hMSCs were also quantified using ex vivo MPI (Fig. 6A), which detected 10,614 ± 1004 cells in the brain, 7212 ± 820 in the lungs, and 92,140 ± 10,437 in the liver. The total calculated distribution numbers of 135,000 (in vivo MPI) and 110,000 (ex vivo MPI) are in close agreement with the total amount of injected cells, being 120,000 (Table 1).

Fig. 5. Dynamic in vivo MPI cytometry comparison of labeled hMSCs and hNPCs after intravenous injection.

Shown are the results for (A) hMSCs and (B) hNPCs. Top panels show PB staining to confirm intracellular iron accumulation in each cell type. Middle panels display time-lapse MPI scans at multiple time points postinjection. Fiducials F1 and F2 contain 10,000 and 20,000 labeled hMSCs in (A) and 20,000 and 40,000 labeled hNPCs in (B), respectively, from the same batch as injected cells. The calibration equations used were as follows: y = 0.1868x + 220 (R² = 0.99) for hMSCs and y = 0.0021x − 2 (R² = 0.99) for hNPCs. Bottom panels present quantitative plots of MPI signal intensity (arbitrary units) and calculated cell numbers over time in each region, demonstrating the ability of MPI for both fast localization and dynamic quantification of labeled cells. Data represent three independent experiments.

Fig. 6. Ex vivo MPI and histology validate cell localization in major organs.

Representative ex vivo MPI and histological analysis of ferucarbotran-labeled hMSCs (A) and Synomag-D70–labeled hNPCs (B) 6 hours after intra-arterial injection. Colocalization of positive PB and anti-HuNa staining confirms the presence of SPIO-labeled hMSCs in the brain.

Table 1. In vivo and ex vivo MPI cytometry of SPIO-labeled hMSCs and hNPCs.

Parameter	hMSCs	hNPCs
Injected cell dose	120,000	960,000
Cell size (μm)	~25	~10
In vivo MPI (12 min postinjection)
Brain (mean ± SD)	22,569 ± 1805	74,240 ± 6854
Lung + liver (mean ± SD)	112,808 ± 10,600	474,580 ± 39,094
Total detected cells	135,377	548,820
% of injected cells detected	113%	57.2%
Ex vivo MPI (6 hours postinjection)
Brain (mean ± SD)	10,614 ± 1004	3,420 ± 280
Lungs (mean ± SD)	7212 ± 820	Not detected
Liver (mean ± SD)	92,140 ± 10,437	143,540 ± 12,430
Lung + liver (mean ± SD)	99,352 ± 10,719	143,540 ± 12,430
Total detected cells	109,966	146,960
% of injected cells detected	91.6%	15.3%

For the smaller hNPCs (Fig. 5B), in vivo MPI demonstrated a dynamic signal evolution in the same anatomical regions. At 12 min postinjection, ~57% of injected hNPCs were detected in vivo, including 74,240 ± 6854 hNPCs in the brain and 474,580 ± 39,094 cells in the lung and liver, corresponding to a (lung + liver)/brain ratio of ~6.5, i.e., ~30% higher than the corresponding ratio observed for hMSCs. The ~43% discrepancy between the injected and detected hNPCs may be attributed to several factors, including diffuse biodistribution below the MPI detection threshold and the inherently lower iron loading of hNPCs relative to hMSCs. In addition, the limited spatial resolution and spillover effect in MPI, particularly from high-signal regions such as the liver, may obscure weak or widely dispersed signals, further contributing to underestimation. Ex vivo MPI (Fig. 6B) detected only 3420 ± 280 cells in the brain, no cells in the lungs, and 143,540 ± 12,430 cells in the liver (Table 1). Histological analysis with PB staining and anti-HuNa immunolabeling only confirmed the presence of SPIO-labeled human cells in the hMSC-injected brains. As hNPCs are much smaller than hMSCs, this could have resulted from a larger washout during postmortem processing compared to hMSCs. Although MPI revealed substantial accumulation of labeled hMSCs in the liver and lungs as well as accumulation of labeled hNPCs in the liver, histological analyses failed to corroborate the corresponding MPI signals. This apparent discrepancy may be attributed to multiple technical and biological factors. First, MPI offers high sensitivity and full-organ detection capability, while histology is inherently limited to thin tissue sections and may miss sparsely distributed cell populations. Second, cells trapped in the microvasculature, especially within the brain, may be lost during organ removal and perfusion, leading to false negatives in postmortem histological analysis. Similarly, in the liver and lungs, cells may localize transiently within sinusoidal spaces or vascular compartments and be washed out during fixation and sectioning.

Splenic localization of hMSCs in the EAE disease model

After establishing the robustness of our MPI cytometry approach in normal mice, we applied the method to an inflammatory setting to evaluate its utility for quantitative in vivo cell tracking under disease conditions. To this end, we performed biodistribution studies in EAE mice as a model of multiple sclerosis (MS) and dynamically tracked 500,000 intravenously delivered ferucarbotran-labeled hMSCs (Fig. 7). We selected hMSCs for these experiments because they are among the most widely used therapeutic cell types in preclinical (33) and clinical (34) MS studies and are known to exert potent immunomodulatory effects (35). Fast dynamic MPI following four sequential intravenous injections (Fig. 7A) revealed that labeled hMSCs were initially retained predominantly in the pulmonary and hepatic compartments. Within hours, however, a distinct MPI signal emerged in the spleen (Fig. 7B), indicating redistribution of a fraction of the injected cells to secondary lymphoid tissue. By 1 day, MPI hotspots were detected in the lungs, liver, and spleen, whereas by 1 week postinjection, only hepatic and splenic signals remained. Ex vivo MPI corroborated these findings, quantifying 61,450 ± 8230 hMSCs in the spleen (Fig. 7C). Histological analysis further validated both in vivo and ex vivo MPI findings. PB staining (Fig. 7C) showed discrete iron-positive foci distributed primarily within the white pulp (leukocyte) region, although, to some extent, also in the red pulp. Anti-HuNa immunofluorescence staining (Fig. 7C) confirmed colocalization of HuNa-positive cells in these areas, confirming that most of the SPIO particles originated from hMSCs instead of endogenous iron present in the red pulp and/or murine macrophages that had phagocytosed SPIO from MSCs.

Fig. 7. In vivo and ex vivo MPI cytometry of intravenously injected ferucarbotran-labeled hMSCs in EAE mice.

(A) Representative real-time in vivo MPI data obtained following four sequential intravenous injections, totaling 500,000 labeled hMSCs. Serial 2D MPI acquisitions (0 to 12 min) reveal rapid signal buildup in the lung/liver region, enabling minute-scale cytometry. Quantification was performed using the fiducial-based regression method (y = 0.10x + 641.8, R² = 0.98), described above for the normal mice. Error bars in the graph represent the SD of MPI signal intensity. (B) Longitudinal tracking of the same mouse using 2D MPI (top) and 3D MPI/CT (bottom) at 2 hours (day 0), 1 day, and 1 week postinjection. A distinct MPI hotspot appears in the spleen (white arrows; S, spleen) beginning at 2 hours and persists through later time points, indicating hMSC redistribution from the pulmonary/hepatic compartments to secondary lymphoid tissue. (C) Ex vivo MPI of the spleen harvested from the same animal, confirming an in vivo splenic signal. PB staining (top right) shows iron-positive foci colocalizing with HuNa-positive cells (bottom right), validating the presence of SPIO-loaded hMSCs within the splenic immunoactive zone of the white pulp. Color bars in each panel show the MPI signal intensity (a.u.).

DISCUSSION

Route of delivery: Intra-arterial versus intravenous cell injection

The findings in this study show how the route of cell delivery largely influences the initial biodistribution and subsequent homing. Intra-arterial injection of SPIO-labeled cells led to a considerably different organ distribution compared to intravenous injection. Consistent with numerous prior imaging studies, intravenously injected hMSCs underwent first-pass sequestration in the lung vasculature (26, 36). This well-known “pulmonary trap” is attributed to the large size (18 to 30 μm in diameter) and adhesive properties of hMSCs, which cause 50 to 80% of intravenously injected cells to lodge in the pulmonary capillaries (36). Our MPI data are in alignment as we observed a strong MPI signal concentrated in the lungs 30 min after intravenous injection (Fig. 2B). This mirrors previous reports using both optical imaging and nuclear imaging. Bioluminescence imaging of luciferase-expressing hMSCs shows a bright lung signal within minutes of intravenous injection, which diminishes over 1 or 2 days as cells clear or die (36). Similarly, SPECT and PET tracking of radiolabeled hMSCs has documented that intravenously injected cells accumulate first in lungs (36, 37), with a subsequent redistribution in the liver (38). Consequently, only a small fraction of hMSCs can reach certain targeted tissues after intravenous injection. Delivering cells via a targeted arterial route allows a greater proportion of injected cells to reach the intended organ. In our study, intra-arterial injection greatly enhanced cell delivery to the brain, whereas an intravenous injection with an approximately eightfold higher dose yielded no detectable brain MPI signal. This observation aligns with prior reports using other modalities. T2-weighted MRI demonstrated that intra-arterial transplantation of hMSCs led to markedly enhanced cell homing in the brain of stroked rats, while intravenous injection of the same cells produced negligible brain localization (39). However, intra-arterial injection of hMSCs may carry a risk of microvascular blockage (39), and hence, its safety must be carefully considered in terms of cell size and cell dose (40). Nevertheless, the ability of intra-arterial administration to bypass the lung “barrier” and achieve near-whole-body distribution is a clear advantage for targeting organs beyond the pulmonary circuit, as demonstrated for targeted renal delivery of hMSCs to the kidney using the renal artery (41) or bone marrow stem cells to the pancreas and spleen using the superior pancreaticoduodenal artery and splenic artery, respectively (42).

By combining dynamic tracking, cell quantification, and disease-model validation, MPI cytometry can elucidate mechanistic principles such as cell size–dependent entrapment, delivery route–specific organ targeting, and disease-driven redistribution, which are difficult or impossible to obtain with conventional imaging modalities. These insights highlight the conceptual breadth of in vivo MPI cytometry and its relevance for guiding therapeutic cell delivery strategies.

Cell size differences: hMSCs versus hNPCs

Beyond the delivery route, the intrinsic properties of the cells led to marked differences in biodistribution and organ retainment (Fig. 5). A robust brain signal could only be observed after injection of 480,000 hNPCs. A major determinant of the cell type–dependent biodistribution is cell size. hMSCs average ~25 μm in diameter (with reported ranges up to 31 μm) (43), whereas hNPCs measure ~9 to 12 μm. Because both cell types exceed the 5- to 8-μm diameter of brain capillaries, their initial localization after intra-arterial delivery is governed primarily by passive, size-dependent microvascular trapping. This mechanical first-pass entrapment explains the stronger cerebral retention of hMSCs compared to the smaller hNPCs.

MPI versus other conventional tracking modalities

MPI enables a level of cell tracking sensitivity and quantification that addresses many limitations of established imaging modalities (MRI, PET/SPECT, and optical imaging). A major advantage of MPI is the ability to quantitatively track small cell populations in deep tissue with high specificity (Fig. 3), as there is no background signal. Hence, there is no ambiguity in identifying SPIO-labeled cells (14, 26). This specificity contrasts with conventional MRI of SPIO-labeled cells, which relies on negative contrast that can be confounded by endogenous sources (e.g., air/tissue interfaces, hemorrhage, or calcifications). Hence, linear quantitation of iron using MPI enabled us to convert the image signal directly into cell number estimates, which is not feasible with MRI. As a result, we could “noninvasively” count the number of cells, i.e., perform “in vivo cytometry” in each organ over time. We are able to detect and quantify cell numbers in the order of thousands with MPI in vivo, which is in line with published sensitivity benchmarks. A recent direct comparison showed that MPI can detect as few as 4000 hMSCs (~75 ng of iron) in a mouse, whereas ¹⁹F MRI may be required on the order of 250,000 cells for detection (17). Another substantial advantage of MPI is its capacity for its dynamic imaging. We leveraged fast scanning and the lack of tissue attenuation in MPI to monitor cell biodistribution immediately after injection and then repeatedly over days to weeks. Such data are difficult to obtain with other modalities. Optical imaging cannot be used in deep organs with tissue attenuation, while ¹⁹F MRI requires lengthy scans. Acquiring just eight contiguous 5.0- by 2.8-cm (2-mm-thick) slices with ¹⁹F MRI takes ~2 hours without whole-body coverage (16). MPI enables rapid whole-body imaging of mice, requiring only 1 to 2 min for two-dimensional (2D) scans, depending on imaging parameters such as scan mode. Recent work has demonstrated real-time “fluoroscopy” with MPI, tracking labeled cells moving through a flow system at subsecond time frame rates (44). Our studies did not require subsecond time frames, but we did acquire serial whole-body snapshots immediately postinjection to observe the first-pass distribution. The ability to see cells in near real time opens the door to optimizing injection techniques (e.g., slow versus bolus injection, use of magnetic targeting, etc.) in ways that static end point measurements cannot.

Although MPI currently achieves a practical spatial resolution of ~1 mm (45), which can be enhanced to ≤300 μm with optimized SPIO tracers and pulse sequences (46), its current resolution remains comparable to that of preclinical PET (~1 to 2 mm) (47) and ¹⁹F MRI (~0.8 mm) (16). However, it still lags behind the sub–100 μm resolution achievable with high-field conventional MRI. The limited spatial resolution of MPI can be effectively mitigated by co-registering MPI with high-resolution MRI (Fig. 3A), allowing the superimposition of quantitative MPI-based cell tracking onto detailed anatomical maps for accurate and spatially resolved cytometry.

Together, our experiments establish MPI as a robust in vivo cytometry platform capable of dynamic, organ-specific quantification of SPIO-labeled cells in deep tissues on a timescale of minutes and across multiple days, capabilities not demonstrated in prior MPI cell tracking studies. MPI provides a highly sensitive, background-free detection of labeled cells but lacks intrinsic anatomical context. Co-registration with CT contributes high-resolution whole-body anatomy, enabling precise assignment of MPI hotspots to specific organs during dynamic tracking, while MRI adds high soft-tissue contrast for validating localization in the brain and other complex structures. This trimodal MPI/CT/MRI framework strengthens the biological interpretation of cell biodistribution by combining quantitative MPI detection with spatial and anatomical fidelity. Although histological PB staining and HuNa staining confirm the presence of human, iron-labeled cells, histology cannot provide whole-organ quantitative validation because it samples only thin 2D sections and may lose intravascular or weakly adherent cells during tissue processing. It requires thousands of sections, which is impossible to quantify especially when the cell distribution is inhomogeneous. Histology also detects intact cells rather than total iron mass, making it inherently incompatible with the quantitative MPI signal. For these reasons, ex vivo MPI, which offers full-organ, background-free quantification, serves as the appropriate validation method for in vivo MPI cytometry and was performed as such in this study.

MPI cytometry in EAE as a neuroinflammatory disease model

When we applied MPI cytometry approach in the EAE disease model, we identified a rapid redistribution of intravenously delivered hMSCs from the lungs and liver to the spleen within hours (Fig. 7). This constitutes in vivo quantitative visualization of splenic hMSC localization in EAE using a noninvasive imaging modality, holding an important mechanistic implication. Although hMSCs rarely accumulate in the central nervous system in sufficient quantity to exert direct local effects, their therapeutic benefit in EAE and MS is widely believed to arise from immunomodulation within peripheral immune organs, especially the spleen (30–32). Our observations are consistent with prior mechanistic studies showing that after intravenous administration, hMSCs can lodge within splenic immune niches and modulate systemic immunity by interacting with T cell and macrophage populations to shift cytokine profiles and suppress pathogenic inflammation in EAE mice (32).

MPI cytometry provided direct evidence that a substantial fraction of hMSCs (calculated as ~12% of injected cells in our study) reaches splenic immune niches, where they may modulate T cell and macrophage populations (48). Quantifying this trafficking pattern may open the door to correlating splenic stem cell load with disease severity, cytokine milieu, and therapeutic outcomes. The power of MPI cytometry lies in its ability to detect and count labeled cells in whole organs, including deep or highly vascular tissues where histology often underestimates the true cell presence because of sampling limitations or postmortem washout. Real-time MPI cytometry can inform on optimal dosing, timing, and delivery strategies for stem cell therapy while being able to correlate variabilities in therapeutic outcome to differences in biodistribution patterns.

Clinical implications and translational outlook

Our study demonstrates a tracking platform with translational potential for integration with (stem) cell therapy. First, the tracer we used (SPIO) is intrinsically suitable for human use. SPIO agents such as ferucarbotran, formerly used as the MRI contrast agent Resovist and now reintroduced as Resotran (49), have well-documented safety profiles in patients, and they can be used to label therapeutic cells without impairing their function (23). The concentration of iron per cell achieved here (tens of pg Fe per cell) is on par with that used in prior human MRI cell-tracking trials, indicating that our labeling approach is scalable to clinical cell doses. Moreover, MPI itself is advancing toward clinical utilization. MPI uses low-frequency magnetic fields that are far below US Food and Drug Administration limits and pose no known risk to tissues (50). Recent engineering breakthroughs have produced human-scale MPI scanners (18–21), and first-in-man studies are now being conducted. As a result, the combination of a nontoxic clinically approved tracer and a safe imaging modality positions MPI-based cell tracking well for rapid clinical translation.

MATERIALS AND METHODS

Study design

This study was designed to evaluate the feasibility, reproducibility and quantitative capability of in vivo MPI cytometry for whole-body tracking of two SPIO-labeled stem cell types (hMSCs and hNPCs) having different sizes. Cells were injected into immunodeficient Rag2^−/− or EAE mice using either intra-arterial or intravenous injection. Dynamic and longitudinal MPI scans were performed to assess whole-body distribution and organ-level cell quantification using fiducial-based calibration for standardization. MRI or CT was performed for anatomical referencing, and the imaging data were verified using histological assessment.

hMSC labeling

P2 human bone marrow–derived MSCs were obtained from RoosterBio and expanded in culture up to P5 using MSC basal medium (Lonza) supplemented with 10% MSC growth supplement, 2% l-glutamine, 0.1% gentamicin, and 0.1% amphotericin. SPIO-labeling conditions were selected on the basis of previous studies demonstrating no adverse effect on hMSC viability, migration, differentiation, or immunomodulatory potency when cells contain in the order of 10 to 30 pg iron per cell (23, 51, 52). To this end, hMSCs were incubated for 24 hours with a pre-prepared labeling medium consisting of MSC basal medium supplemented with ferucarbotran (25 μg Fe/ml, Meito Sangyo Co.) and poly-l-lysine (PLL; Sigma-Aldrich, P-1524, 3 μg/ml). Following incubation, the labeling medium was removed, and the labeled hMSCs were washed three times with phosphate-buffered saline (PBS) to remove unbound particles. A total of 120,000 hMSCs were administered per intra-arterial injection, while 500,000 hMSCs were administered per intravenous injection.

hNPC labeling

Induced pluripotent stem cells (iPSCs) derived from a patient with amyotrophic lateral sclerosis and genetically corrected for the SOD1 mutation (39B2.5 line) were obtained from K. Eggan (53). iPSCs were differentiated into NPCs using a defined neural induction medium supplemented with dual-SMAD inhibitors (54). Briefly, iPSCs were plated on Matrigel-coated plates and cultured to confluence in Essential 8 medium (Gibco). Neural induction was initiated using WiCell neural induction medium supplemented with dual-SMAD pathway inhibitors including LDN193189 (Stemgent) and SB431542 (Sigma-Aldrich). Starting on day 3, retinoic acid (Sigma-Aldrich) was added to promote caudalization of the neuroectoderm. On day 5, the medium was replaced with neural induction medium containing retinoic acid and continued until day 12. From day 12 to day 25, cells were maintained in neural induction medium supplemented with 1 μM purmorphamine (Sigma-Aldrich), brain-derived neurotrophic factor (10 ng/ml; Invitrogen), glial cell line–derived neurotrophic factor (10 ng/ml; R&D Systems), ciliary neurotrophic factor (10 ng/ml; R&D Systems), 200 μM ascorbic acid (Sigma-Aldrich), and insulin-like growth factor 1 (10 ng/ml; R&D Systems) to support NPC maturation. Cells were harvested on day 25 for subsequent labeling with Synomag-D70.

For hNPCs, the labeling procedure was performed in a standard 24-well tissue culture plate. Each well contained 100,000 hNPCs suspended in 2 ml of complete growth medium. Cells were labeled with Synomag-D70 (40 μg Fe/ml), and PLL (390 ng/ml). Cells were incubated for 4 hours at 37°C. After incubation, cells were washed three times with 10 mM PBS (pH 7.2 to 7.4).

Extensive optimization was required to establish a robust labeling protocol for hNPCs because of their smaller size and lower endocytic volume compared to hMSCs. We evaluated multiple SPIO formulations (ferucarbotran, feraheme, Synomag-D50, and Synomag-D70), several transfection agents (PLL, protamine sulfate, and polybrene), different culture plates, varied incubation durations (2 to 72 hours), and physical delivery methods including electroporation and mechanoporation. None of these conditions yielded efficient intracellular iron uptake. Synomag-D70 combined with PLL was the only condition that consistently produced high labeling efficiency while maintaining cell viability (>90%), and this optimized protocol was therefore used for all hNPC experiments in this study.

Cell viability, PB staining, and ferrozine iron assay

To verify that SPIO labeling did not affect cell viability, we assessed hMSCs and hNPCs immediately after labeling and before each injection using an acridine orange/propidium iodide live/dead viability kit (Logos Biosystems) and a LUNA dual-fluorescence automated cell counter. Labeled cells were mixed with an acridine orange/propidium iodide staining solution, loaded into a LUNA counting slide, and analyzed according to the manufacturer’s protocol. Automated quantification of green (live) and red (dead) cells consistently showed >90% viability for both cell types under the labeling conditions used in this study. PB staining was performed to detect the presence of iron within labeled cells. Cells were fixed with 4% glutaraldehyde for 20 min, washed with PBS, and incubated with Perl’s reagent (potassium ferrocyanide in HCl) for 30 min at room temperature. To enhance cellular visualization, cells were counterstained with nuclear fast red (NFR) for 20 min. Stained cells were washed with deionized water to remove excess reagent and then imaged using a Zeiss Apotome 2 microscope. The Fe content in labeled cells was determined using a ferrozine assay, as described elsewhere (55). A standard curve prepared with known Fe concentrations was used to calculate the Fe content in the sample.

In vivo injection procedures

All animal procedures were approved by the Johns Hopkins University Animal Care and Use Committee (protocol/approval number: MO24M118). Male immunodeficient Rag2^−/− mice (6 to 8 weeks old, the Jackson Laboratory) or female C57BL/6 mice (4 to 6 weeks old, the Jackson Laboratory) were housed under a 12-hour light/dark cycle with ad libitum access to food and water. All injection, surgery, and in vivo imaging procedures were performed under anesthesia using 5% induction and 1.5% maintenance isoflurane. Before injection, all labeled cell suspensions were assessed for iron content and labeling efficiency. For intra-arterial injection, the ICA (internal carotid artery) was surgically catheterized as previously described (40, 56), and cells were delivered using a Harvard Apparatus infusion pump. Intravenous injections were performed via a tail vein catheter and a 27G needle using the same pump system. EAE was induced in C57BL/6 mice using a MOG_35-55/CFA Emulsion kit (EK-0111, Hooke Laboratories). Fourteen days after EAE induction, mice were catheterized in the tail vein for intravenous injection (n = 3). Cells were suspended in 200 μl of sterile PBS and delivered intra-arterially or intravenously either as a single injection or in four sequential injections at 3-min intervals for dynamic imaging studies. The infusion rate was maintained at 0.15 ml/min. Following intra-arterial injection, animals received a single subcutaneous dose of extended-release buprenorphine (1 mg/kg) for postoperative analgesia.

The hMSC and hNPC total doses (960,000 hNPCs versus 120,000 hMSCs) were not standardized because the two cell types differ markedly in iron-loading capacity. hMSCs efficiently internalize SPIO, whereas hNPCs load substantially less iron even under optimized Synomag-D70/PLL conditions. Standardizing the injected cell number would therefore result in vastly different total iron doses and would place hNPCs below the MPI detection threshold. Instead, we adjusted the hNPC dose to achieve a comparable detectable iron mass.

MPI

MPI was performed using a Momentum field-free line scanner (Magnetic Insight Inc.). A custom mouse bed with two lateral racks, each featuring holes to hold fiducials, was designed in FreeCAD (version 0.21.0) and fabricated using a 3D printer (Ultimaker 2 Extended+). For in vivo imaging, a 2D whole-body MPI scan was first acquired in standard mode, followed by a 3D scan (also in standard mode) using 21 projections. The standard mode was selected for its rapid acquisition speed, making it well suited for dynamic in vivo imaging (57). Before use, the imaging bed was scanned to confirm the absence of magnetic contaminants. Two fiducials containing different numbers of SPIO-labeled cells (from the same batch as those injected) suspended in 100 μl of PBS were positioned within the field of view (FOV) for both MPI signal quantification and image co-registration. Regions of interest (ROIs) were manually delineated on “sharpened” MPI images using ImageJ. Windowing parameters (brightness, contrast, minimum, and maximum) were consistently adjusted across datasets acquired under the same conditions to minimize blooming artifacts, particularly in areas with high signal intensity (e.g., liver). The MPI signal intensity for each hotspot was calculated by multiplying the mean signal intensity by the ROI area.

Micro-CT

Using the same in vivo MPI bed equipped with a custom 3D-printed adaptor for co-registration, mice were maintained under anesthesia and promptly transferred to an adjacent IVIS Spectrum/CT system (Caliper Sciences) to acquire anatomical micro-CT images. Micro-CT scans were performed with a tube voltage of 50 kVp (kilovolt peak), a current of 200 μA, and an exposure time of 300 ms. Images were acquired at a voxel resolution of 100 μm, and 3D reconstructions were generated using 3D Slicer software.

MRI

High-resolution MRI was performed ex vivo only. To enable co-registration of MPI and MRI data for the mouse head, a custom MPI/MRI cradle was designed using FreeCAD (version 0.21.0) and fabricated with a 3D printer (Ultimaker 2 Extended+). The cradle included two alignment disks, each with a hole to hold fiducial tubes for spatial referencing. Following MPI acquisition of the excised mouse head mounted in a 3D-printed cradle, high-resolution ex vivo MRI was performed using a Bruker BioSpec 17.6T vertical bore scanner equipped with a Bruker BioSpin 25-mm proton transmit/receive volume coil and ParaVision 5.1 software. A 3D FLASH gradient-echo sequence was used with the following parameters: echo time = 2.6 ms, repetition time = 8.4 ms, flip angle = 5°, FOV = 20 mm by 30 mm by 18 mm, and matrix size = 200 × 300 × 180, resulting in an isotropic spatial resolution of 100 μm. To improve the signal-to-noise ratio, 16 signal averages were acquired.

MPI/MRI and MPI/CT data co-registration

Co-registration was performed using the Volume, Look-Up Table, Transform, and Rendering tools in 3D Slicer. Two fiducial tubes containing labeled cells were included in the imaging setup and used as reference points for the accurate alignment of MPI and MRI/CT datasets. When comparing MPI with MRI and CT, we did not observe major discrepancies in signal localization. However, each modality carries inherent detection biases. To minimize the MPI blooming artifact from the strong liver signal, sharpening and windowing parameters were kept constant across scans. As MRI hypointensities can be confounded by endogenous susceptibility sources, only MRI voids that colocalized with MPI hotspots and were confirmed by PB and anti-HuNa staining were interpreted as true positives. CT was performed solely to provide anatomical reference and does not contribute to SPIO detection. In ambiguous cases, ex vivo MPI provided a background-free gold standard for detection, allowing us to distinguish the true signal from false positives or false negatives across modalities.

Histopathology

Following in vivo imaging, mice were euthanized and transcardially perfused first with 1× PBS, followed by 4% paraformaldehyde before the brain, lungs, heart, and liver were collected for ex vivo imaging and postmortem analysis. Tissues were immersed in 4% paraformaldehyde at 4°C for 24 hours and then transferred to 30% sucrose for another 72 hours. After embedding in optimal cutting temperature compound, the organs were sequentially cut into 20-μm sections using a cryostat (Thermo Fisher Scientific). For PB staining, tissues were fixed with 4% glutaraldehyde for 20 min and stained as described above. Anti-HuNa staining was performed to visualize hMSCs or hNPCs. Tissue sections were first blocked with 5% bovine serum albumin and 0.1% Triton X-100 in tris-buffered saline (1× TBS; pH 7.0) for 2 hours, followed by incubation with mouse anti-HuNa (1:250, NBP2-34342, Novus-Bio). Goat anti-mouse Alexa Fluor 488 (1:500, A-11001, Invitrogen) was used as a secondary antibody, prepared in 3% bovine serum albumin in 1× TBS. Sections were incubated for 2 hours at room temperature and then triple washed in 1× TBS to remove the unbound antibody in each step. Sections were coverslipped with mounting medium containing 4′,6-diamidino-2-phenylindole. Fluorescence microscopy was performed using a Zeiss Axiovert 200 M inverted epifluorescence microscope.

Statistical analysis

Linear regression analysis was performed to plot the total MPI signal [arbitrary units (a.u.)] for each fiducial against Fe content or cell quantity. The R² value of each regression line was used to evaluate linearity. Data are expressed as the means ± standard deviation (SD) from at least three independent experiments (technical replicates). All statistical analysis and data visualization were performed using Microsoft Excel.