Cold Quad-Modal Nanocomplex for Precise and Quantitative In Vivo Stem Cell Tracking

Abstract

Current single imaging modalities typically lack the ability to simultaneously offer detailed anatomical visualization and quantitative cellular information, which is crucial for evaluating and improving therapeutic efficacy. This work develops a cold (non-radioactive) quad-modal imaging nanocomplex for magnetic resonance imaging (MRI), magnetic particle imaging (MPI), computed tomography (CT), and multispectral optoacoustic tomography (MSOT) within a single nanoplatform, that is, Albumin-Bismuth-Superparamagnetic iron oxide or ABS. The chemically engineered complex is composed of bovine serum albumin as biocompatible matrix, superparamagnetic iron oxide as MRI and MPI agents, and optoradiopaque bismuth sulfide as CT and MSOT agents. This work demonstrates here its first use for high-resolution, real-time, and quantitative in vivo imaging of mesenchymal stem cells transplanted in mouse brain. This versatile nanocomplex may find applications for non-invasive monitoring cell transfer and cell transplantation in vivo using multiple imaging approaches.

1. Main

Further advances in stem cell therapy and regenerative medicine will benefit from in vivo imaging techniques for cell localization and quantification.^[1–4] Numerous approaches for in vivo tracking of stem cells have been used in preclinical and clinical studies, including the use of positron emission tomography (PET),^[5] single photon emission tomography (SPECT),^[6] bioluminescence imaging (BLI),^[7] computed tomography (CT),^[8] magnetic resonance imaging (MRI),^[9,10] magnetic particle imaging (MPI),^[11,12] and multispectral optoacoustic tomography (MSOT).^[8] Each imaging modality has its own strengths and weaknesses for tracking cells in vivo (see Table 1). Multimodal imaging may be used to combine strengths and obviate weaknesses when there is a simultaneous, complementary reporting on cell biodistribution, quantity and viability with high resolution, specificity, and sensitivity.^[13] The effective use of multimodal imaging hinges on specific clinical or preclinical aims, and is dependent on multiple single imaging agents or ideally a multimodal imaging probe including those based on nanoparticles (NPs).^[14,15] Hence, further development of versatile NPs that can be detected with multiple imaging modalities is warranted.^[16] Several studies have developed tri-modal,^[17,18] and quad-modal,^[19,20] agents to enhance the imaging information across multiple modalities. However, most reported multimodal NPs rely on incorporating “hot (radioactive)” agents for quantitative tracking using PET and SPECT,^[19–21] and require bespoke manufacturing and radiation safety processes. Advancements in MPI and MSOT technologies have recently helped overcome such limitations. MPI is sensitive, specific, and quantitative,^[11] but it lacks anatomical information and an excellent spatial resolution for cell tracking, while MSOT offers quantification,^[22]spectral selectivity and high spatiotemporal resolution.^[23] Hence, the development of “cold” multimodal NPs that respond to MRI and CT as anatomical imaging modalities, and MPI and MSOT as quantitative methods will offer opportunities to paint a comprehensive picture of in vivo cell distribution.

Table 1.

Comparison of the different clinical imaging modalities that are available for in vivo tracking of stem cells. Red: Modalities that use “hot” tracers. Blue: Modalities that use “cold” tracers or contrast agents used as quad-modal agents in this study.

Criteria	PET	SPECT	US	CT	MRI	MPI	MSOT
Ionizing radiation	Yes	Yes	No	Yes	No	No	No
Sensitivity	Excellent	Excellent	Excellent	Limited	Moderate	Excellent	Excellent
Specificity	Excellent	Excellent	Excellent	Limited	Limited	Excellent	Moderate
Quantification	Moderate	Moderate	Limited	Excellent	Limited	Excellent	Moderate
Anatomical content	No	No	Moderate	Excellent	Excellent	No	Moderate
Spatial resolution	Limited	Limited	Moderate	Excellent	Excellent	Moderate	Moderate
Temporal resolution	Limited	Limited	Excellent	Excellent	Excellent	Excellent or Moderate	Excellent

We report here on a “cold” versatile nanocomplex for quad-modal cell tracking that combines superparamagnetic iron oxide (SPIO) NPs and optoradiopaque bismuth sulfide (Bi₂S₃) NPs within a bovine serum albumin (BSA) matrix. This integration leverages the magnetic properties of SPIO for MRI^[10] and MPI,^[11] with the dual optical and radiopaque characteristics of Bi₂S₃ providing contrast for CT^[24] and MSOT.^[25] The BSA matrix enhances the biocompatibility and stability of the nanocomplex,^[26] ensuring its integrity as a single platform suitable for comprehensive multimodal imaging.

1.1. Nanocomplex Synthesis and Physicochemical Characterization

Using a step-by-step solvothermal decomposition method, albumin-bismuth (AB) nanocomplexes composed of bovine serum albumin (BSA) and Bi₂S₃ were synthesized first, followed by the addition of SPIO to form albumin-bismuth-SPIO (ABS). Ferucarbotran (a carboxydextran-coated iron oxide and the active pharmaceutical ingredient in Resovist) was chosen as SPIO for the ABS nanocomplex due to its commercial availability (ensuring quality control and widespread use^[27]) and excellent properties for both MRI^[28] and MPI.^[29,30] While Resovist was discontinued by Bayer-Schering Pharma (Berlin, Germany), ferucarbotran is still manufactured in Japan by Meito-Sangyo Co. Ldt. (Nagoya, Japan), relabeled and diluted by Magnetic Insight Inc. as Vivo-trax. In addition, Bayer-Schering has now rebranded Resovist as Resotran,^[31,32] aimed at its use as clinical MPI tracer. Bi₂S₃ is a better opto-radiopaque agent than the more commonly used gold NPs, as it shows a narrower bandgap and a larger X-ray attenuation coefficient (1.3 versus 5 eV and 5.8 versus 5.1 cm⁻² kg⁻¹ at 100 KeV, for Bi₂S₃ and gold NPs, respectively).^[33]

The chemical composition, functional groups, and optical properties of ABS were characterized by Fourier transform infrared (FTIR) and ultraviolet-visible (UV–vis) spectroscopy. Surface charge, size, particle concentration, morphology, composition, and protein binding assessments of ABS were performed by zeta potential analysis, dynamic light scattering (DLS), nanoparticle tracking analysis (NTA), transmission electron microscopy (TEM), high-angle annular dark field (HAADF), inductively coupled plasma mass spectrometry (ICP-MS), multi-angle light scattering (MALS), and asymmetric-flow field-flow fractionation (AF4). Key findings are summarized in Figure 1, while complementary characterization results are provided in Figures S1–S6 and Tables S1–S11, Supporting Information.

Figure 1.

Physicochemical characterization of ABS. A) Schematic illustration and appearance of ABS. B) FTIR spectra comparing BSA, ferucarbotran, AB, and ABS. C) Particle size distribution for AB and ABS measured using AF4 coupled with MALS and DLS detectors to determine protein binding to their surface. D) UV–vis absorbance spectra and size distribution histograms of ferucarbotran, AB, and ABS. E) TEM images of AB and ABS at different magnifications. F) HAADF STEM and bright-field STEM images and elemental mapping showing the nanoscopic distribution profile of Fe, S, and Bi.

Figure 1A depicts the structure and appearance of ABS in DI water as a dark near-black formulation. FTIR spectra show the BSA peaks at 1648 and 1540 cm⁻¹, characteristic of amide I (C═O stretching vibrations) and amide II bands (N─H bending and C─N stretching), respectively (Figure 1B). The peak in the BSA spectrum just below 3000 cm⁻¹ represents CH₂ stretching vibrations. These vibrations across the spectra of BSA, AB, and ABS indicates that BSA is indeed integrated with the AB and ABS, and that the addition of SPIO and Bi₂S₃ NPs does not significantly alter the overall structure of the BSA matrix. In the ferucarbotran and ABS spectra, the Fe-O stretching observed between 600–550 cm⁻¹ corresponds to the SPIO core. The spectral peaks ≈3400 cm⁻¹ for O—H stretching, 1750–1700 cm⁻¹ for C=O stretching, and 1150 cm⁻¹ for C—O stretching are indicative of the carboxydextran coating of the SPIO core.

Particle size distributions for AB and ABS were measured using AF4 coupled with MALS and DLS detectors. Samples were also incubated with human plasma to assess surface protein binding. AB had one major peak with an averaged hydrodynamic diameter of 84 nm which increased to 99 nm after plasma incubation (Figure 1C). The corresponding radius of gyration sizes for AB were 52 and 79 nm, respectively. This resulted in a shape factor (2*Rg/Dh) of 1.23 without plasma incubation and 1.59 after plasma incubation, implying protein binding to the surface of the AB. Similar to AB, ABS also had one major peak with an averaged hydrodynamic diameter of 86 nm. Following plasma incubation, the size increased to 102 nm (Figure 1C). The corresponding radius of gyration sizes for ABS were 45 and 75 nm, respectively, resulting in a shape factor of 1.04 without plasma and 1.47 after plasma incubation, demonstrating that there was a similar protein binding to the surface of ABS. Incubation of AB and ABR with human plasma led to an increased hydrodynamic diameter and altered shape factors, which are both hallmarks of protein corona formation which is known to enhance the cellular uptake by stem cells.^[34]

UV–vis spectra demonstrate that the absorbance of ferucarbotran decreases with increasing wavelength, a feature typical for metal oxide NPs^[35] (Figure 1D). However, for AB and ABS which incorporate metal sulfide NPs, different optical properties exist due to their unique electronic structures and different light interactions.^[36] The variations in the absorption profiles for ABS compared to AB and ferucarbotran likely result from the combined effects of SPIO and Bi₂S₃, illustrating the complex interaction of these components within the ABS nanocomplex. DLS analysis revealed that ferucarbotran exhibits a peak particle size of 8.5 nm and a Z-average of 40.6 nm (Figure 1D). AB displays a peak size of 75.6 nm with a Z-average of 104.3 nm, while ABS shows a peak size of 104.5 nm with a Z-average of 135.1 nm. Further characterization data on particle sizing, concentration assessment, and quantitative elemental analysis are provided in Figures S1–S5, Tables S1–S11, Supporting Information.

TEM images indicate that both AB and ABS possess a spherical morphology (Figure 1E). ABS is characterized by larger clusters, likely arising from the incorporation of SPIO. ABS was also characterized using HAADF STEM, bright field STEM, and elemental analysis to determine its morphology and composition at high-definition (Figure 1F). The findings demonstrate the integration of multi-component NPs within the ABS formulation, highlighting a uniform physical structure/size and good dispersibility, as well as the effective encapsulation and distribution of SPIO and Bi₂S₃ NPs within the BSA.

Full zeta potential analysis of AB and ABS at native pH of 7.29 showed negative values of −23.1 ± 1.5 and −24.3 ± 1.0 mV, respectively, while ferucarbotran was neutral at native pH of 7.29 (Figure S6, Table S11, Supporting Information). The conjugation of BSA to Bi₂S₃ in the AB nanocomplexes most likely originates from a combination of coordination binding and surface adsorption by functional groups on both the protein and the Bi₂S₃ NPs. Previous studies on BSA-stabilized Bi₂S₃ NPs have shown that BSA can directly participate in the nucleation and stabilization process by binding to exposed Bi sites during particle growth.^[37] Under the alkaline conditions used in our solvothermal synthesis, it is probable that Bi₂S₃ surfaces present exposed Bi³⁺ sites or partially coordinated sulfide centers, which can act as Lewis acid (electron-accepting) centers capable of coordinating electron-donor groups on BSA (e.g., thiols, amines, carboxylates). Previous studies of Bi₂S₃ NP ligand coordination (e.g., Sn-doped Bi₂S₃ showing ligand binding of Bi atoms) support this interpretation.^[38] Moreover, ligand-mediated growth and surface functionalization in Bi₂S₃ systems have been shown to depend heavily on the reactivity of surface Bi atoms toward coordinating functional groups.^[39] Electrostatic attraction likely further promotes BSA adsorption to Bi³⁺ centers on Bi₂S₃ surfaces, consistent with strong Bi³⁺ binding to albumin observed in prior studies.^[40] At physiological or slightly alkaline pH, BSA carries a net negative charge,^[40,41] which can interact favorably with partially positive charges on surface Bi atoms. This dual interaction (coordination and electrostatic binding) facilitates the formation of a stable protein shell around the Bi₂S₃ core. In addition, the flexible tertiary structure of BSA enables partial conformational rearrangement, allowing the protein to “wrap” around Bi₂S₃ NP surfaces and fill nanoscale voids to enhance steric stabilization, as shown in studies of BSA-coated NP systems where protein structural adaptation contributes to colloidal stability.^[42] The negative zeta potential measured for ABS (Table S11, Supporting Information), compared with the neutral ferucarbotran formulation, is consistent with the BSA shell dominating the surface chemistry. This suggests that the dextran or iron oxide coating is embedded beneath the BSA layer. This implies that the BSA shell effectively masks the native surface charges of both Bi₂S₃ and SPIO components while maintaining strong binding interactions beneath the coating.

To enhance the cell detection sensitivity of CT to that of MRI and MPI, which are inherently more sensitive, we designed ABS with an initial Bi to Fe mass ratio of 5:1. This ratio was intended to enhance CT detectability when the same probe is used for cell labeling. The actual concentrations of Bi and Fe in ABS were quantified using ICP-MS, confirming an experimental ratio aligning with the theoretical ratio (Table S8, Supporting Information).

1.2. Quantitative In Vitro Imaging Studies of Naked ABS

The imaging properties of ferucarbotran, AB, and ABS are shown in Figure 2. Quantification of CT contrast in Hounsfield Units (HU) revealed no contrast enhancement for ferucarbotran, but AB and ABS demonstrated a linear increase in CT contrast with concentration (R² = 0.98–1.00, Figure 2A). T₂-weighted (w) MRI showed that there is no such relation in signal intensity for ABS, inherent to the non-quantitative nature of MRI for detecting SPIO. We observed no differences in MRI signal intensity between ABS and ferucarbotran for the same Fe concentration, nor did AB show notable changes compared to water at equivalent Bi concentrations. To assess the NMR transverse relaxivity, multi-echo T₂ mapping was performed to calculate r₂ values for both ABS and ferucarbotran. Linear regression of R₂ = 1/T₂ versus [Fe] yielded r₂ values of 47 mM Fe⁻¹s⁻¹ for ABS and 310 mM Fe⁻¹s⁻¹ for ferucarbotran at 11.7 T, respectively (Figure S7, Supporting Information). The reduced T₂-shortening efficiency of ABS indicates partial magnetic shielding of the SPIO cores within the BSA–Bi₂S₃ matrix. The slightly higher r₂ value measured for ferucarbotran at 11.7 T (309.8 s⁻¹ mM⁻¹) compared to the reported value at 9.4 T (267.6 s⁻¹ mM⁻¹, most relevant available r₂ data for ferucarbotran at high magnetic fields^[43]) is likely due to greater local field inhomogeneities at higher magnetic field strengths.

Figure 2.

Quantitative phantom imaging studies. Two different concentration categories, including high (1 mg Fe/mL or 5 mg Bi/mL) and low (100 μg Fe/mL or 500 μg Bi/mL) concentrations were tested. A) CT imaging of serial dilutions of ferucarbotran, AB, and ABS formulations with corresponding HU values versus concentration. B) MRI and signal intensity analysis of ABS. The highest equivalent concentration per Fe for ferucarbotran and AB were also imaged with MRI under the same conditions and are shown as the control group for ABS. C) MPI comparing different concentrations of ferucarbotran, AB, and ABS formulations. D) MSOT of ABS, ferucarbotran, and AB formulations. The highest equivalent concentration per Bi for ferucarbotran and AB were also imaged by MSOT and shown as the control group for ABS. “Ref” in panel D contained 100 μL of DI water. Data in each panel represent three independent experiments.

As expected, AB was undetectable by MPI due to the absence of SPIO, whereas ABS and ferucarbotran exhibited a strong MPI signal (Figure 2C). The MPI performance of ABS was slightly higher than ferucarbotran, suggesting underlying differences in magnetic relaxation that warranted further investigation. To this end, relaxation measurements were performed using the RELAX software module on our MPI scanner (Figure S8, Supporting Information). Both ABS and ferucarbotran tracers exhibited symmetric peaks centered at ≈0 mT, characteristic of superparamagnetic NP behavior. However, at the same iron mass concentration, the ABS sample showed a broader and higher peak amplitude compared to ferucarbotran, indicating faster magnetic relaxation dynamics and higher magnetic susceptibility. This broader response suggests that the addition of Bi₂S₃ and albumin in ABS modifies the microenvironment around the SPIO core, potentially enhancing Néel relaxation and signal strength, which can contribute to improved sensitivity in MPI.^[44] Vibrating sample magnetometry (VSM) further confirmed the magnetic differences between ABS and ferucarbotran. The two tracers exhibited similar hysteresis profiles (Figure S9, Supporting Information), indicating superparamagnetic behavior. However, the saturation magnetization of ABS was lower than ferucarbotran when both were normalized by mass, suggesting reduced magnetic content or altered magnetic coupling within the ABS nanocomplex.

Although ferucarbotran contains more iron per unit of total mass and exhibits a stronger saturation magnetization (Figure S9, Supporting Information), MPI relaxometry (Figure S8, Supporting Information) and imaging (Figure 2C) showed that the ABS is a more sensitive MPI tracer. This apparent contradiction arises because VSM measures static magnetization at high external magnetic fields, whereas MPI signal generation depends primarily on the nonlinear dynamic magnetization response at kHz frequencies near zero field. As reported by Tay et al., static VSM is not predictive of MPI performance because temporal relaxation effects dominate the signal.^[45] In ABS, the presence of a large fraction of non-magnetic Bi₂S₃ relative to iron lowers the overall magnetic content and consequently reduces the saturation magnetization measured by VSM. However, the incorporation of Bi₂S₃ in the composite formulation can modify the local microenvironment of the SPIOs, reducing interparticle interactions and effective anisotropy. This leads to faster Néel relaxation and enhanced dynamic magnetic response at the MPI drive frequency, enabling stronger MPI signals despite the lower iron content. A similar behavior was observed by Arami et al., who demonstrated that particles with faster relaxation generate stronger MPI signals even if their static magnetization is lower.^[44] Conversely, the higher iron content and denser magnetic core aggregation in ferucarbotran (compared to ABS) promote Brownian-dominated relaxation, resulting in slower magnetization dynamics under physiological conditions.^[46] This frequency-dependent attenuation of MPI signal has also been confirmed by Moor et al.^[47] Therefore, ABS could outperform ferucarbotran as an MPI tracer possibly because its faster Néel relaxation enables stronger harmonic generation and a more efficient dynamic response at MPI-relevant frequencies.

In addition to CT and MPI, MSOT also proved to be quantitative, as evidenced by the linear response (R = 0.93) of signal versus concentration (Figure 2D). No differences in MSOT signal intensity were observed between ABS and AB at equivalent Bi concentrations. Similarly, ferucarbotran at the same Fe concentration as ABS showed no change in MSOT signal intensity compared to water.

Thus, except for MRI, the amount of ABS can be quantified with CT, MPI, and MSOT as they exhibit a linear response with concentration. The minimum detectable ABS concentration for MSOT was ≈12.5 μg Fe mL⁻¹ or 62.5 μg Bi mL⁻¹ (Figure 2D), which is two times higher than that for MPI (Figure 2C). Given a Bi:Fe ratio of 5:1, suggests that the MPI sensitivity for Fe detection is effectively 10 times higher than the MSOT sensitivity for Bi detection. Based on clinical protocols, an HU of greater than 50 is considered an acceptable minimum for contrast enhancement of CT imaging,^[48] the minimum detectable ABS concentration can be considered at 100 μg Fe mL⁻¹ or 500 μg Bi mL⁻¹ (Figure 2A). Consequently, the CT detection sensitivity is 8 times lower than MSOT for Bi detection.

1.3. Cellular Uptake, Cytotoxicity, and In Vitro Imaging of ABS-hMSCs

We have chosen hMSCs as a cellular prototype for ABS labeling, reflecting their extensive clinical use for treating a wide range of diseases including ischemic heart disease,^[49] osteoarthritis,^[50] diabetes mellitus,^[51] neurological disorders,^[52] and cancer,^[53] with over 1000 registered clinical trials highlighting their therapeutic relevance. Figure 3A illustrates the cellular uptake and peri-nuclear localization of ABS, its parental components, ferucarbotran and AB after Prussian Blue (PB) staining. TEM was used to further assess labeling at the ultrastructural level (Figure 3B). A Ferrozine spectrophotometrical assay for iron quantification revealed an average intracellular content of 92.95 ± 6.72 pg of Fe per single cell.

Figure 3.

Cellular uptake of ABS. A) PB staining of hMSCs labeled for 24 h with 25 μg Fe/mL ABS and its parental components ferucarbotran and AB. B) TEM images of cells labeled with ferucarbotran, AB, and ABS at various magnifications, showing the intracellular distribution and aggregation of NPs within subcellular structures.

To evaluate the absence of cytotoxicity following labeling, hMSCs were incubated with different concentrations of ferucarbotran, AB, and ABS formulations for 24 or 72 h. While ferucarbotran exhibited no significant cytotoxic effects across all tested concentrations and time points, ABS started to display cytotoxicity at concentrations ≥150 μg Bi mL⁻¹ after 24 h and ≥100 μg Bi mL⁻¹ after 72 h incubation (Figure 4A). We therefor selected a concentration of 25 μg Fe mL⁻¹ equivalent to 125 μg Bi mL⁻¹ and 24 h as the working concentration and incubation time for all further experiments. An (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay confirmed >95% cell viability immediately after labeling under these conditions, demonstrating no acute cytotoxicity. To further assess long-term biocompatibility and proliferative potential, a 2-week follow-up experiment combining PB staining and an acridine orange/propidium iodide (AO/PI) live/dead cell assay showed >90% viability with sustained label retention across two passages (Figure S10, Supporting Information).

Figure 4.

Cell viability and phantom imaging studies of ABS-labeled hMSCs. Two different cell densities were tested, 10000 cells/μL and 1000 cells/μL. A) Cell viability of ABS-labeled hMSCs for varying concentrations of Fe or Bi and incubation times. *p < 0.05 compared to control (0, unlabeled) group. B) CT, C) MPI/MRI, and D) MSOT of serial dilutions of ABS-labeled hMSCs with corresponding quantification and linear regression analysis. “Ref” in panel D contained 100 μL of 8% gelatin with 0.02% w/v sodium azide added. Data in each panel figure represent three independent experiments.

ABS-labeled hMSCs were imaged at two different cell densities, 10 000 and 1000 cells μL⁻¹ using CT (Figure 4B), MPI and MRI (Figure 4C), and MSOT (Figure 4D). Unlike MRI signal, the quantitative nature of CT, MPI, and MSOT was affirmed when R² values exceeded 0.9. CT contrast did not notably enhance at cell densities below1000 cells μL⁻¹, where HU was lower than 50. Both MPI and MSOT successfully detected a minimum cell density of ≈62 ABS-labeled hMSCs per μL. This does however not imply that MPI and MSOT have the same sensitivity for cell detection, given that the Bi:Fe ratio used was 5:1. Hence, MPI exhibits the highest cell detection sensitivity among the four imaging modalities.

Multi-echo T₂ mapping at cell densities lower than 1000 cells μL⁻¹ demonstrated a strong linear correlation between the transverse relaxation rate (R₂ = 1/T₂) and the density of ABS-labeled hMSCs (Figure S11, Supporting Information), confirming that MRI signal loss scales linearly with increasing cell density (slope: 0.14 s⁻¹ · μL cell⁻¹, R² = 0.983). This result validates the quantitative relationship between SPIO content of ABS and R₂ enhancement.

We employed an 11.7 T small-animal MRI scanner primarily to maximize spatial resolution and signal-to-noise, thereby enabling rigorous evaluation of our ABS-labeled hMSCs in a preclinical mouse model. Although clinical MRI systems operate at lower field strengths, prior work has demonstrated that cells labeled with SPIO agents such as ferucarbotran can generate visible contrast even at 1.5 T^[54] and 3 T.^[55] Thus, the choice of 11.7 T in the present work reflects a deliberate preclinical experimental design decision rather than direct translation to clinical scanners. We recognize that translation to human-scale MRI at 3 T (or 1.5 T) will likely entail reduced contrast and lower relaxivity changes due to weaker susceptibility effects and lower magnetic field strength. Importantly, current preclinical experimental design was to allow comparison with the other preclinical (MPI, CT and MSOT) imaging modalities. While the MSOT system used here is applicable to both animal and clinical research, commercially available human-scale MPI scanners are still under development and not yet widely accessible.

1.4. In Vivo Imaging of ABS-Labeled hMSCs Transplanted in Mouse Brain

Immunodeficient Rag2^−/− mice were used to prevent immediate immune rejection of transplanted hMSCs. Figure 5A shows bi-modal in vivo MPI and MRI conducted at various time points after cell injection depicting the ability to track the spatial localization and persistence of ABS-hMSCs over time simultaneously. Both magnetic imaging modalities did exhibit blooming artifacts, with its actual size not notably changing with varying cell densities on MRI. In contrast, the size decreased with reduced cell density on MPI. Ex vivo CT and MSOT imaging were also conducted one-month post-injection, accompanied by histological analysis (Figure 5A) using PB staining and anti-human nuclear antigen (HuNA) staining to validate the imaging findings, with a good agreement with the imaging data. All four imaging modalities were able to detect cell densities as low as 6250 ABS-hMSCs per μL. The observation for CT, despite being the least sensitive method tested, is in agreement with the in vitro detection threshold being 1000 ABS-hMSCs per μL (Figure 4B).

Figure 5.

Serial in vivo and ex vivo imaging and quantitative analysis of ABS-labeled hMSCs injected intracerebrally in normal Rag2^−/− mice (n = 4). A) In vivo MPI/MRI of mice injected with varying numbers of ABS-labeled hMSCs at different time points post-injection. M1, M2, M3, and M4 received 1.00 × 10⁵, 5.00 × 10⁴, 2.50 × 10⁴, and 1.25 × 10⁴ cells, respectively. Ex vivo CT and MSOT imaging, along with histology of M1 (as a representative example), were also performed to further validate ABS-labeled hMSC distribution and localization. B) Time-course analysis of in vivo MPI signal intensity, showing a signal decay over 30 days for different cell quantities. C–E) Correlation between in vivo MPI, ex vivo CT, and ex vivo MSOT signal intensity and the number of injected cells, demonstrating strong linear relationships.

A time-course analysis of MPI signal intensity demonstrated a gradual signal decay over 30 days (Figure 5B). This decline may result from progressive label dilution at the injection site,^[56] cell death,^[57] and/or exocytosis^[58] of the ABS nanocomplex, followed by phagocytic clearance by host macrophages.^[59] As this was a proof-of-concept study performed in normal, non-diseased Rag2^−/− mice, our primary objective was to evaluate the multimodal imaging capabilities of ABS-labeled hMSCs rather than to dissect these biological mechanisms in detail. However, in vivo MPI data (Figure 5A,B) indicate that the gradual signal decay observed over 30 days is more consistent with progressive label clearance from the injection site than with acute cell death or immediate immune uptake. As shown in Figure 5A, PB staining and anti-HuNA immunostaining one month post-injection demonstrate a strong spatial co-localization between the iron signal and human cell nuclei. Based on this finding, it is reasonable to assume that similar or even higher levels of co-localization were present at earlier time points. This observation indicates that the majority of the iron label remained associated with injected hMSCs, suggesting minimal transfer of the label to host cells within this period in a normal brain microenvironment.

Clinical MRI, CT, and MSOT scanners are already available, while MPI technology is progressing toward clinical translation with human head-sized scanners^[60–65] and handheld devices.^[66] Figure 5C–E presents the quantitative correlation between the number of ABS-hMSCs injected and the corresponding signal intensities obtained from MPI, CT, and MSOT, respectively. Each imaging modality demonstrated a strong, dose-dependent linear relationship between signal intensity and cell number, confirming their ability to accurately quantify labeled cell populations in vivo. Among these, MPI exhibited the highest sensitivity for cell detection (Figure 4), enabling robust detection and quantification of small cell populations, while CT and MSOT provided complementary spatial and anatomical information.

2. Conclusions

We introduced ABS as a nanocomplex for multimodal anatomical and quantitative imaging of stem cells, leveraging the unique properties of albumin, Bi₂S₃ NPs, and ferucarbotran. Through comprehensive physicochemical characterization, we established the robust stability of ABS in terms of structural integrity and functional performance. Key findings include the preservation of structural components of albumin within the ABS nanocomplex and the incorporation of SPIO and Bi₂S₃ without altering the primary BSA protein structure. We found that CT, MPI, and MSOT have distinct sensitivity thresholds for quantifying both naked ABS and ABS-labeled cells. MPI demonstrated the greatest sensitivity, followed by MSOT and then CT, while MRI signal was found not to be quantitative.

Our results highlight the ABS capability to address some of the existing limitations of MRI, CT, MPI, and MSOT. The integration of MPI and MRI provided a dual approach to visualize cells in a quantitative manner within their anatomical context, a significant advancement over single-modality magnetic imaging. With interventional radiology cell-injection procedures most commonly performed using CT or fluoroscopy, the inclusion of Bi₂S₃ in ABS not only enhances the sensitivity of CT but also enables sensitive real-time stem cell tracking using MSOT.

Looking forward, ABS may present numerous opportunities to develop new imaging-guided precision cell therapies. Integrating ABS with advanced technologies such as targeted drug delivery systems and real-time monitoring of therapeutic interventions could significantly advance cell therapy.

3. Experimental Section

Materials for Synthesis of AB and ABS:

Ferucarbotran (stock concentration 1 M Fe) was purchased from Meito Sangyo. BSA (Cat. # A9647, Lot # SLCH8436, pH 7, ≥98% purity), bismuth (III) nitrate pentahydrate (Bi(NO₃)₃.5H₂O), 70% nitric acid, and NaOH were purchased from Sigma-Aldrich, and 1-ethyl-3-(3ʹ-dimethylaminopropyl)carbodiimide, hydrochloride (EDAC, HCl) or C₈H₁₇N₃ · xHCl was purchased from EMD Millipore Corp. Lot variations of BSA were observed to result in slight size variations of AB and ABS, but all data were acquired with high consistency using the specific BSA lot listed above.

Synthesis of AB:

Three solutions were prepared. Solution 1 was prepared by dissolving 13.2 g of BSA in 200 mL of double-distilled water (ddH₂O) in a glass container under continuous stirring until all BSA granules were completely dissolved. Solution 2 was prepared by dissolving 17 mg of Bi(NO₃)₃·5H₂O in 2.88 mL of 70% HNO₃ in a 15 mL tube. After the Bi salt was completely dissolved, 11.12 mL of ddH₂O was added, and the mixture was shaken to achieve a homogeneous solution. Solution 3 was prepared by dissolving 6 g of NaOH in 30 mL of ddH₂O in a 50 mL tube at pH = 12.75). While Solution 1 was being stirred vigorously in the glass container, Solution 2 was added dropwise. If white precipitation appeared during this process, the solution was homogenized using an ultrasonic bath to ensure a fully transparent yellow solution. Finally, Solution 3 was added to the mixture of Solutions 1 and 2 while stirring vigorously. After 30 mins of stirring, the color of the mixture changed from light yellow to black. The mixture was then stirred gently for an additional 8 h. The final product was collected by centrifuging the mixture at 13,500×g for 10 min, followed by washing the precipitate three times with ddH₂O. The yield was ≈87%, based on bismuth content.

Synthesis of ABS:

Six mg of EDAC was mixed with 2 mL of SPIO (Ferucarbotran, 1 mg Fe mL⁻¹), followed by the addition of 12 mL of ddH₂O to bring the total volume to 14 mL. The mixture was shaken for 15 min. Subsequently, 2 mL of AB (5 mg Bi mL⁻¹) was added, and the mixture was shaken for an additional 2 h. The final product was washed three times with ddH₂O using centrifugation at 6400×g for 5 min, employing a centrifugal filter tube (Amicon Ultra-15, 10 kDa MWCO) for purification. The collected ABS nanocomplexes were then sonicated for 15 min. The yield was ≈90%, based on Fe content.

Physicochemical Characterization of ABS:

A Shimadzu UV-1900i spectrophotometer and a Shimadzu FT-IR 4300 instrument were used for optical characterization. A Malvern Zetasizer Nano ZS instrument with back scattering detector (173°) was used for measuring the hydrodynamic size and zeta potential in batch mode. The following NIST-NCL joint protocol PCC-1 was followed: https://www.cancer.gov/nano/research/ncl/protocols-capabilities. Stock samples were diluted 100-fold in water, 10 mM NaCl, and PBS. Samples were measured at 25 °C in a quartz microcuvette. Zeta potential measurements were made at native pH and, where necessary, after adjustment to near neutral pH (7.4) using either 1 N standardized HCl or 1 N standardized NaOH. Sample pH was measured and/or adjusted before loading into a pre-rinsed folded capillary cell.

Particle size distribution was also measured using AF4 coupled with MALS and DLS detectors. The AF4 system consisted of an isocratic pump (Agilent G1310A), well-plate autosampler (Agilent G1329A), UV–vis detector (Agilent G1315B), AF4 separation channel (DualTec, Wyatt Technology), MALS detector (Wyatt HELEOS II), refractive index (RI) detector (Wyatt OptiLab T-rEX), and a DLS (Malvern Zetasizer Nano ZS) instrument. The separation channel had a length of 275 mm and a 350 μm spacer. A 10 kDa MWCO regenerated cellulose membrane was used for particle separation. The detector flow was 1 mL min⁻¹ and the injection volume was 100 μL for all samples. The mobile phase was PBS. The elution cross-flow was 1.50 mL min⁻¹ for 10 min, followed by a linear decrease to 0 mL min⁻¹ in 3 min, and held at 0 mL min⁻¹ for 25 min. The membrane was passivated with 5 mg mL⁻¹ BSA for a total of three runs prior to sample analysis. The chromatographic traces were monitored by absorption at 210 nm, MALS, and DLS detection. MALS normalization constants were determined using BSA at 5 mg mL⁻¹ in PBS. DLS was used for measuring the hydrodynamic diameter in flow-mode. Samples were diluted 100-fold in PBS prior to injection. For plasma incubation experiments, 5 μL of sample were incubated with 100 μL of human plasma at 37 °C for 2 h with agitation. Then, 895 μL PBS was added to the mixture, which was further diluted 1:1 with PBS for a final 400-fold dilution in 5% human plasma.

The hydrodynamic size and particle concentration were also measured using a Wyatt DynaPro plate reader III (Wyatt Technology). The instrument was calibrated with dextran standards and the PBS solvent was filtered through a 0.02 μm regenerated cellulose membrane prior to use. The size and particles per mL concentration were also measured using NTA (View-Sizer 3000). Measurements utilized all laser lines. Samples were diluted in water accordingly (100 000 to 2 000 000-fold) to give a particle count of ≈107 particles mL⁻¹.

The iron, sulfur and bismuth concentrations in ABS and AB were determined by ICP-MS. A Perkin-Elmer NexION 2000B equipped with a micromist nebulizer, standard sample introduction system, and integrated autosampler operated in standard mode was used. Data was analyzed using the Syngistics software. Samples for Bi and Fe analysis were prepared by serial dilution with three dilution steps. Dilution 1:50 μL of sample was weighed out and digested in 400 μL HNO₃ and 100 μL of HCl, followed by dilution to 10 mL using a 2% HNO₃ solution. Dilution 2:100 μL of Dilution 1 was weighed out and diluted to 10 mL using 2% HNO₃. Dilution 3:1 mL of Dilution 2 was weighed and diluted to 10 mL using 2% HNO₃. Total weights of the diluted solutions were also recorded and used to determine the dilution factors and calculate the amount of Fe or Bi in solution. Samples for sulfur analysis were prepared by single dilution. Dilution 1:50 μL of sample was weighed out and diluted in 10 mL of Milli-Q atwer. Iron calibration standards were prepared from a serial dilution of a 0.997 mg g⁻¹ Fe in HNO₃ (Inorganic Ventures). The iron calibration standards prepared had concentrations of 0, 2.186, 4.367, 6.845, 8.668, and 11.10 ng g⁻¹. Bismuth calibration standards were prepared from a serial dilution of a 0.997 mg g⁻¹ Bi in HNO₃ (Inorganic Ventures). The bismuth calibration standards prepared had concentrations of 0, 10.40, 20.65, 30.85, 41.28, and 51.81 ng g⁻¹. Sulfur calibration standards were prepared from a serial dilution of a 0.997 mg g⁻¹ S (Inorganic Ventures) in water. The sulfur calibration standards prepared had a concentration of 0, 0.9688, 5.001, 10.13, 30.13, and 50.28 ng g⁻¹.

All concentrations were determined using an external calibration constructed using the standards above. Iron, bismuth, and sulfur were measured using peak hopping with an integration time of 50 ms. Each measurement consisted of 3 sweeps with 6 reading per sweep, and the final Intensity was determined from an average of 10 replicants for each sample. Samples were run in KED mode with He as the reaction gas. Masses analyzed included Bi-209, Fe-57, and S-34.

Labeling of hMSCs:

P2 human bone marrow-derived MSCs were obtained from RoosterBio and expanded in culture up to P5 using MSC basal medium (MSCBM, Lonza) supplemented with 10% MSC growth supplement, 2% l-glutamine, 0.1% gentamicin, and 0.1% amphotericin. For all labeling experiments except the cell viability assay, hMSCs were incubated for 24 h with a pre-prepared labeling medium consisting of MSCBM medium supplemented with either ABS (25 μg Fe mL⁻¹), AB (125 μg Bi mL⁻¹), or ferucarbotran (25 μg Fe mL⁻¹), along with poly-L-lysine (Sigma P-1524) as a cationic agent at a concentration of 3125 ng mL⁻¹. Following incubation, the labeling medium was removed, and the labeled hMSCs were washed three times with PBS to remove unbound particles. TrypLE Express (Gibco) was used to gently detach the labeled cells for subsequent staining or injection into mice.

Prussian Blue Staining:

PB staining was performed to detect the presence of iron within labeled hMSCs. Cells were fixed with 4% glutaraldehyde for 20 mins, 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 for 20 mins. Stained cells were washed with deionized water to remove excess reagent and then imaged using a Zeiss Apotome 2 microscope.

Ferrozine Assay:

The Fe concentration in naked NPs or labeled hMSC was determined using a Ferrozine assay as described elsewhere.^[67] A standard curve prepared with known Fe concentrations was used to calculate the Fe content in the sample.

Cell Viability Assay:

To evaluate the potential cytotoxicity of ferucarbotran, AB, and ABS, a lactate dehydrogenase (LDH) assay (ThermoFisher Scientific) was performed for 24 or 72 h incubation with various NP concentrations. hMSCs were seeded in 96-well plates and following incubation the culture supernatant was collected, and the absorbance was measured at 490 nm according to the manufacturer’s protocol. The cytotoxicity was expressed as percentage of the absorbance relative to the positive control (100% cell death).

hMSC viability following ABS labeling was further assessed using an MTT assay. Cells were seeded in 96-well plates and cultured for 24 h under standard conditions. Cells were then incubated with ABS (25 μg Fe mL⁻¹) complexed with poly-L-lysine (Sigma P-1524, 3125 ng mL⁻¹) in 0.1 mL of culture medium, while untreated cells served as controls. After 24 h of incubation and three times washing with PBS, the medium was replaced with 100 μL of PBS containing 0.5 mg MTT (Sigma-475989) per mL, and cells were incubated for an additional 4 h at 37 °C. The MTT solution was subsequently replaced with 100 μL of dimethyl sulfoxide to solubilize the purple formazan crystals, followed by gentle shaking for 10 min. Absorbance was measured at 570 nm using an ELISA microplate reader (PerkinElmer (Wallac) VICTOR3 1420 Multilabel Counter), and the cell viability of ABS-labeled hMSCs was calculated as a percentage relative to untreated control cells.

TEM:

Nanocomplexes (10 μL) were adsorbed on glow-discharged ultra-thin carbon-coated 400 mesh copper grids (EMS CF400-Cu-UL). Cells were fixed in 4% paraformaldehyde (PFA), 0.1% glutaraldehyde, 3 mM MgCl₂ in and 0.1 M Sorenson’s sodium phosphate buffer, pH = 7.2 overnight at 4 °C. After buffer rinse, samples were postfixed in 1% osmium tetroxide and 1.5% potassium ferrocyanide in 0.1 M sodium phosphate for 1 h on ice in the dark. Samples were rinsed in ddH₂O, dehydrated in a graded series of ethanol and embedded in Eponate 112 (Polyscience) resin, and polymerized at 60 °C overnight. Sixty to 90 nm sections were cut with a diamond knife on a Leica UCT ultramicrotome and collected with Formvar coated 2 × 1 copper slot grids. Before imaging, grids were stained with 2% uranyl acetate, rinsed with ddH₂O and stained with 0.3% Reynolds lead citrate. Grids were imaged with a Hitachi 7600 TEM at 80 kV equipped with an AMT CCD XR80 (8-megapixel camera - side mount AMT XR80 - high-resolution high-speed camera).

HAADF-STEM imaging was performed using a JEOL JEM-ARM200F microscope operated at 200 kV to visualize the morphology and structural features of ABS. For elemental analysis, EDS mapping for Fe, Bi and S was conducted using an Oxford Instruments X-MaxN detector integrated into the microscope.

Labeled Cell Injections:

All animal experiments were approved by our Institutional Animal Care and Use Committee. Male immunodeficient Rag2^−/− mice (n = 4, 6–8 weeks old, The Jackson Laboratory) were housed under a 12-h light/dark cycle with unrestricted access to food and water. Intracerebral injections were performed under 1–2% isoflurane anesthesia with mice secured in a stereotaxic frame (Stoelting Co.). A burr hole was created 0 mm caudal and 2 mm lateral to the bregma. Using a 31G Hamilton syringe (Hamilton), 2 μL of cell suspension in PBS was injected into the striatum at a depth of 3 mm below the endocranium. Varying quantities of cells (1 × 10⁵, 5 × 10⁴, 2.5 × 10⁴, and 1.25 × 10⁴ cells) were delivered at a rate of 0.2 μL min⁻¹ over 10 min, with the needle left in place for 5 mins before withdrawal. The incision was closed using 3–0 vicryl sutures, and 1 mg kg⁻¹ buprenorphine ER was administered s.c. once post-surgery.

Quad-Modal Imaging:

To directly compare the in vitro sensitivity of the four imaging modalities, naked AB, naked ABS and ABS-labeled hMSC sample were prepared at different cell densities. Naked NP samples were suspended in 100 μL of DI water in small 200 μL Eppendorf tubes. Labeled cell samples were prepared in 100 μL of 8% gelatin with 0.02% w/v sodium azide added. In vivo MRI and MPI were performed at 1 h, 1 week, and 1 month after ABS-hMSC injection. Following in vivo MRI/MPI imaging, mice were sacrificed and ex vivo MSOT/CT imaging were conducted, followed by post-mortem analysis.

MRI:

MRI was conducted using a 11.7 T Bruker Biospin horizontal bore scanner equipped with a 25-mm-volume coil and interfaced with ParaVision 6.0.1 software. In vitro MRI was performed to generate 2D T2-w images using a multi-slice multi-echo sequence. In vitro MRI parameters were repetition time (TR) = 2200 ms, echo time (TE) = 6.5 ms, and slice thickness = 1.5 mm. The image matrix size was 192 × 192, with a field of view (FOV) of 40 × 40 mm, resulting in an in-plane resolution of 0.2 × 0.2 mm. Each in vitro scan used one average and one repetition.

A 3D FLASH sequence was employed for in vivo MRI studies, featuring isotropic imaging with a slice thickness of 40 mm and an image matrix size of 200 × 125 × 250. The FOV was 32 × 20 × 40 mm, with a resolution of 0.16 × 0.16 × 0.16 mm. Scan parameters were TE = 3.4 ms, TR = 25 ms, and flip angle = 7° with fat suppression. Each in vivo scan used a single average with one repetition. Regions of interest (ROIs) were manually drawn and data were processed using ImageJ.

MPI:

MPI was performed using a Momentum field-free line imager (Magnetic Insight Inc.). Phantoms were positioned vertically within a 3D custom-printed holder (Ultimaker 2 Extended+), and 2D images were acquired using the manufacturer’s “standard mode.” For samples with lower iron concentrations or cell densities, tubes were imaged individually by placing eppendorf tubes vertically in the center of the holder, aligned with the MPI FOV.

For in vivo imaging, a 2D whole-body MPI scan was initially performed in standard mode, followed by 3D MPI scan (standard mode) with 21 projections ROIs were manually drawn and data were processed using ImageJ.

MPI/MRI Data Co-Registration:

Co-registration was performed using the Volume, Look-Up Table, Transform, and Rendering tools in 3D Slicer. To facilitate co-registration and quantification, two fiducial tubes containing ABS-labeled hMSCs, prepared from the same batch of injected cells, were included in the imaging setup. These fiducials were then used as reference points for accurate alignment of MPI and MRI datasets.

MSOT:

MSOT imaging was performed using an MSOT inVision 512-echo scanner (Version 4.0.2.0, iThera Medical, GmbH, Munich, Germany), equipped with a 512- toroidally-focused ultrasound transducer array operating at a central frequency of 5 MHz, spanning a circular arc of 270° to effectively detect optoacoustic signals. Light excitation was provided with a tunable optical parametric oscillator pumped by an Nd:YAG laser. Excitation pulses with a duration of 9 ns at wavelengths ranging from 660 to 900 nm at a repetition rate of 10 Hz, wavelength tuning speed of 10 ms, and a peak pulse energy of 100 mJ at 720 nm were used. Sample tubes were embedded in cylindrical phantoms.^[68,69]

For ex vivo MSOT, mice were first transcardially perfused with 10 mM PBS and 4% PFA. Heads were collected, shaved, and wrapped in a thin polyethylene membrane and then placed in a customized animal holder (iThera Medical). A thin layer of ultrasound gel was applied to the skin for optimal acoustic coupling to the membrane prior to imaging. MSOT data were acquired with excitation wavelengths ranging from 660 to 900 nm in steps of 5 nm, with 10 frames recorded and averaged for each wavelength. All MSOT data analysis was performed using viewMSOT software (Version 4.0.2.0, iThera Medical, GmbH, Munich, Germany). A linear regression method was used to perform multispectral processing. MSOT images were reconstructed from the raw data using a back-projection algorithm at a resolution of 100 μm. ROIs were manually drawn based on concurrently acquired b-Mode ultrasound images. Within each ROI, the mean of the highest 10% of pixels was used for quantification. Analyses were computed with a NP unmixing preset, using the spectrum derived from a sample with highest NP concentration.

Micro-CT:

An IVIS Spectrum/CT imaging system (Caliper Sciences) was used with a tube voltage of 50 kVp, a current of 200 μA, and an exposure time of 300 ms. Images were acquired at a voxel size of 100 μm, and 3D reconstructions were generated using the accompanying software. Images were processed using ImageJ. To quantify CT signal intensity as HU, a two-point calibration method was used. The CT signal intensity in water was set to 0 HU, and that for ambient air to −1000 HU. Sample HU values were then obtained by linear extrapolation using

$$HU=1000\times \frac{{\mu }_x-{\mu }_{\mathit{\text{water}}}}{{\mu }_{\mathit{\text{water}}}-{\mu }_{air}}$$ (1)

where, μ_x, μwater, and μ_air are the linear attenuation coefficients of the sample, water, and air, respectively.

Histopathology:

After ex vivo imaging, the brains were removed and immersed into 4% PFA at 4 °C for 24 h, and then transferred to 30% sucrose for another 72 h. After embedding in optimal cutting temperature compound, the brains 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 staining was performed as described above. Anti-human nuclear antigen (anti-HuNA) staining was performed to visualize hMSCs. Tissue sections were first blocked with 5% BSA and 0.1% Triton X-100 in Tris-buffered saline (1X TBS, pH = 7.0) for 2 h, 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 secondary antibody, prepared in 3% BSA in 1X TBS. Sections were incubated for 2 h at room temperature and then triple washed in 1X TBS to remove antibody for each step. Sections were cover-slipped with mounting medium containing DAPI. Fluorescence microscopy was performed using a Zeiss Axiovert 200 M inverted epifluorescence microscope.

Statistical Analysis:

Linear regression analysis was performed to plot the total signal (a.u.) from each imaging modality against NP concentration or labeled cell density. The R² value of each regression line was used to evaluate the linearity between the signal and either NP concentration or cell density. Data are expressed as the mean ± standard deviation from at least three independent experiments (technical replicates). Statistical analysis was conducted using GraphPad Prism software (version 10.3.0). Comparisons across multiple groups were carried out using one-way analysis of variance followed by Tukey’s post hoc test, with a significance level of p < 0.05.

Supplementary Material

Supporting Information is available from the Wiley Online Library or from the author.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.