Optical fluorescence imaging is frequently applied to noninvasively track the biodistribution of fluorophore-labeled drugs, nanoparticles, and antibodies.1–5 Commonly used fluorophores are near-infrared dyes such as heptamethine cyanine dyes. Planar fluorescence imaging methods are the most established techniques, but they do not consider absorption and scattering of organs and tissues and lack geometric information, making quantitative analysis difficult and prone to error. Quantitative methods that do not require anatomical reference modalities are ex vivo approaches such as cryofluorescence tomography or cryofluorescence micro-optical sectioning tomography.6–8 Such destructive methods render 3-dimensional (3D) fluorescence images by successively slicing the object and capturing reflectance images of the surface, which are then registered spatially with microscopic white light images.

In contrast, tomographic methods such as fluorescence tomography (FLT) reconstruct 3D images based on the diffuse light distribution and can therefore also be applied in vivo. After a migrating laser excites the 3D object at different positions, quantitative FLT reconstructs the 3D fluorescence distribution from the raw optical data, which may include the estimation and use of scattering and absorption maps (Fig. 1D).9,10 To support the automated segmentation of scattering maps, most commercially available FLT scanners, such as the MILabs micro–computed tomography (CT) optical imaging system (MILabs B.V., Houten, the Netherlands), the IVIS Spectrum CT (PerkinElmer, Waltham, MA), the AMI HTX (Spectral Instruments Imaging, Tucson, AZ), and the InSyTe FLECT-CT (TriFoil Imaging, Chatsworth, CA), are hybridized with an integrated CT or can be combined with a separate CT such as the FMT 2500 LX (PerkinElmer). In addition, the use of CT allows organ segmentation required for analysis based on anatomical information, which increases reproducibility, compared with interpreting it into the fluorescence data.11 However, the soft tissue contrast of CT is limited, and thus segmenting soft tissues for scattering maps and assessing organ biodistribution might be inaccurate. Contrast agents potentially compensate for the considerably low soft tissue contrast in CT and thus increase the accuracy of FLT. However, these can interact with metabolism and elimination,12 potentially affecting the biodistribution of the optical probe. In addition, exposure to ionizing radiation can have an impact on animals' welfare as well as on tumor physiology.13

FIGURE 1:

CT-MRI fusion and fluorescence tomography reconstruction. Various markers were tested to enable fusion (A). The mouse bed holder is shown with the inner markers for CT-MRI fusion, outer markers for CT-FLT fusion, and 1 side marker for orientation in the mouse holder frame (blue) (B). For CT-MRI fusion, MRI scans were corrected for gradient nonlinearity distortion and chemical shift using a LEGO phantom filled with water (outer part, in magenta) and coconut oil (inner part, in turquoise) (C). After CT-MRI fusion, various scattering segmentations for FLT reconstructions were tested (D). The state-of-the-art CT-FLT reconstruction considered fat, bone, lungs, skin (560 μm), and unspecified soft tissue. The CT-FLT reconstruction with extended scattering information considered additionally scattering parameters of the liver and kidneys (CT-FLT*). MRI-FLT reconstruction considered fat, lungs, skin (280 μm), and unspecified soft tissue. On top of that, the MRI-FLT reconstruction with extended scattering information considered the brain, liver, and kidneys (MRI-FLT*), whereas the combined CT-MRI-FLT reconstruction with extended scattering information also considered bones (CT-MRI-FLT*). Based on the scattering segmentation and the measured absorption, scattering and absorption maps were reconstructed to compute the light scattering and absorption for the FLT reconstruction.

Compared with CT, magnetic resonance imaging (MRI) has a better and more versatile soft tissue contrast that can be shaped by the pulse sequences. Thus, hybridizing FLT with MRI is highly attractive, which may enable the inclusion of additional organs. Furthermore, it enables multiple functional and metabolic analyses, for example, by applying arterial spin labeling, functional MRI, blood oxygenation level–dependent imaging, chemical exchange saturation transfer, and diffusion-weighted sequences, as well as MR spectroscopy.14–17

Academic sites already merged FLT with MRI and developed simultaneous and sequential MRI-FLT applications.18–20 Simultaneous MRI-FLT reduces scanning time and has been used for tumor imaging but requires complex engineering to make the FLT compatible with the high magnetic field strengths.19 Sequential MRI-FLT, composed of separated scanners or units, has been used for brain and tumor imaging.18,20 However, no MRI-FLT reconstruction has been presented yet to assess biodistribution on a whole-body scale.

This study sought to answer 3 important questions: (1) “Can a reliable MRI-FLT imaging approach be developed for whole-body imaging?,” (2) “Will the consideration of organ-specific scattering improve the accuracy of FLT reconstruction?,” (3) “Which method, CT-FLT, MRI-FLT, or CT-MRI-FLT, is best for segmentation-based FLT biodistribution analyses?”

MATERIALS AND METHODS CT-FLT Imaging

Fluorescence tomography scans were performed by excitation with a 730-nm laser at various positions in a hybrid micro-CT optical imaging system (MILabs B.V., Houten, the Netherlands), requiring a scan time of approximately 15 minutes. After optical imaging, the mouse holder was automatically moved into the CT unit, where scans were performed in total body normal scan mode, tube voltage of 55 kV, tube current of 0.17 mA, an isotropic voxel size of 140 μm, and a scan time of 4 minutes 11 seconds, resulting in a total scan time of approximately 20 minutes. Computed tomography–FLT fusion and reconstruction are based on the contrast of 44 air-filled outer marker holes in the plastic multimodal mouse holder (animal holder for hybrid CT-FLT imaging, MILabs B.V.). The reconstructed FLT data had an isotropic voxel size of 280 μm.

Marker-Based Co-registration of CT and MRI Data

As neither air nor plastic generates MRI signals, various marker substances were tested to fill the inner marker holes of the mouse holder, comprising water, sucrose, white petrolatum, oil, gelatin, agar, and glycerol (Fig. 1A). A total of 45 petrolatum markers, sealed with transparent nail polish, of approximately 5 mm3 were placed in the mouse holder frame, consisting of 11 markers on each side of the frame and 1 side marker (Fig. 1B). To reduce the MRI's field of view (FOV) and scanning time, the inner marker holes were used.

MRI Sequence Optimization and Correction for Gradient Nonlinearity Distortion and Chemical Shift

The effects of geometric distortions were first evaluated in a LEGO phantom because of its well-defined structure (Fig. 1C). It consisted of 10 adjacent 2 × 2 LEGO bricks (The LEGO Group, Billund, Denmark) arranged in 2 rows, and its volume corresponded to a mouse placed in the multimodal mouse holder (80 × 30 × 9 mm3). The inner and outer compartments of the bricks were filled with coconut oil and water, respectively.

Magnetic resonance imaging was performed using a 7 T BioSpec 70/20 USR MRI scanner (Bruker BioSpin, Ettlingen, Germany) running on ParaVision 7.0.0 (Bruker BioSpin) with an RF Res 300 1H 112/086 QSN TO AD volume coil commonly used for rat imaging with a coverage of 42 mm in z-direction. The Dixon-based T2-weighted fat-water–separated turbo RARE (rapid acquisition with relaxation enhancement) sequence was well suited for whole-body imaging, offering a reasonable scan time and visibility of the white petrolatum marker. The sequence had the following parameters: FOV, 80 × 55 × 17.6 mm3; matrix, 384 (frequency encoding direction) × 264 (phase encoding direction) × 44 (slices); repetition time, 2715.0 milliseconds; effective echo time, 22.6 milliseconds; echo spacing, 11.3 milliseconds; RARE factor 4; echo shift for water-fat separation, 0.37 milliseconds; resolution, 208 μm/pixel in-plane (coronal); slice thickness, 400 μm without slice gaps; receiver bandwidth, 78,125.0 Hz; and a scan time of 11 minutes 56 seconds. For each slice, water and fat images were reconstructed by the scanner software ParaVision.

Measurements on the uniformly shaped LEGO phantom demonstrated that the magnetic field gradient in the FOV was not linear and the spatial position of the sample was incorrectly reconstructed, resulting in a barrel-shaped image of the LEGO phantom. Activating the “gradient linearity correction” in the ParaVision software solved this problem.

In addition, misregistration due to chemical shift caused by imaging 2 adjacent materials with different Larmor frequencies was visible in the LEGO phantom. The predicted shift between fat and water images was 1.1 mm, calculated by multiplying the MRI working frequency ω0 with the chemical shift δ and the FOV and dividing by the receiver bandwidth.21 In our LEGO phantom measurements, we observed a 1.5-mm shift, corresponding to 7 voxels in the MRI matrix. This shift was considered in the adjusted MRI-FLT reconstruction software.

Assessment of the CT-MRI Co-registration

Computed tomography–MRI fusion was evaluated by assessing the agreement of CT- and MRI-aided segmentations of the LEGO phantom via the Dice score.22 Dice score calculations were performed using Imalytics Preclinical 3.0.2.1 (Gremse-IT GmbH, Aachen, Germany) by dividing 2 times the volume of the overlap by the volume of both segmentations (Fig. 2). Dice scores above 0.8 were ranked good, ≤0.8 to 0.6 as medium, and <0.6 as poor.23,24 Furthermore, the residuals of fusing CT and MRI based on markers were considered. The mean distances of the respective CT and MRI markers are a measure of image co-registration. The closer this score gets to 0, the better the markers match.

FIGURE 2:

Organ segmentation and Dice score calculation. Dice scores calculate the overlap between 2 segmentations (A, upper panel). Illustrative, the complete segmentation of CT (A, lower panel blue) and MRI (A, lower panel red) are shown in coronal view. The actual organ segmentations can be found in 2D (B, upper panel for MRI-aided segmentations, lower panel for CT-aided segmentations) and 3-dimensional (3D) (C) with CT on the left and MRI on the right. Bo, bone; Br, brain; Cae, caecum; Co, colon; Gb, gallbladder; H, heart; Ki, kidney; Li, liver; Lu, lungs; Spi, spine; Spl, spleen; St, stomach; ThG, thyroid gland.

Because the MRI sequence should also enable the identification of mouse shape, as well as body fat and organs in addition to marker detection and fusion, we subsequently applied it to dead mice. As for the LEGO phantoms, the agreement of segmentation with CT scans was evaluated with Dice scores.

Fluorescence Reconstruction Considering Information From CT or MRI and Multimodal FLT Reconstruction

Fluorescence tomography reconstructions were performed by taking anatomical reference values from either CT, MRI, or both. In addition, it was investigated whether taking into account extended scattering information is beneficial for the reconstructions. In Figure 1D, screenshots of the various approaches are shown.

CT-FLT Reconstruction

The CT-aided FLT reconstruction (MILabs FLT Recon 1.1.1.8, developed by Gremse-IT GmbH, Aachen, Germany) provided with the MILabs system was based on the approach described by Gremse et al.10 The computations performed were GPU-accelerated, hence a Quadro P5000 (NVIDIA, Santa Clara, CA) with compute capability 6.1 was used, which takes approximately 15 minutes per scan. The marker holes in the bed are automatically detected in the CT and 2D optical raw data to co-register the data and enable accurate reconstruction. In this reconstruction, the scattering map is determined from an automated organ segmentation and fixed scattering coefficients: fat (14.0 cm−1), bone (17.1 cm−1), lungs (20.5 cm−1), skin (25.9 cm−1, segmented 560 μm thick), and unspecified tissue (11.2 cm−1).10

CT-FLT Reconstruction With Extended Organ Scattering Information (CT-FLT*)

The standard scattering map (see CT-FLT reconstruction) was extended by adding scattering coefficients for the liver and the kidney with 8.0 cm−1 and 19.8 cm−1 (at 730 nm), respectively.25

MRI-FLT Reconstruction

The MRI-FLT reconstruction was similar to the CT-FLT reconstruction but adjusted to load the MRI data, automatically detect the markers in the MRI data, and perform an automated organ segmentation. Based on the water and fat images, the FLT reconstruction software created a total of 3 channels: fat, water, and mixed. During the reconstruction, the marker positions in the MRI scans were determined and co-registered with the CT to compare the segmentations. Based on this, MRI, CT, and FLT were fused using Imalytics Preclinical 3.0.2.1 (Gremse-IT GmbH, Aachen, Germany).22 Magnetic resonance imaging measurements showed that the skin (cutis) was approximately 230 μm thick, which was confirmed by histological examinations (200–300 μm).26 Thus, in contrast to the CT-FLT reconstruction, the skin segmentation was adjusted to only 1 voxel layer of 280 μm. Here, the same classes as in the CT-FLT reconstruction were considered, disregarding the bones and adjusting the skin thickness.

MRI-FLT Reconstruction With Extended Organ Scattering Information (MRI-FLT*)

In comparison to the extended CT-FLT reconstruction described above (CT-FLT*), MRI-aided segmentation correctly identified the brain. Thus its scattering coefficient of 12.1 cm−1 (at 730 nm) was additionally considered in the MRI-FLT* reconstruction.25

CT-MRI-FLT Reconstruction With Extended Organ Scattering Information

Like the MRI-FLT* reconstruction, CT-MRI-FLT* reconstruction was performed with extended scattering information while combining the advantages of both modalities. Thus, fat, liver, kidneys, and brain were segmented via MRI, whereas lung and bone information from CT were considered. Total body and skin (280 μm) contours were taken from CT.

Fluorescence Quantification With Fluorescence Inserts

To calibrate the fluorescence signal of the applied dye (from signal intensities to concentrations in pmol), a block-shaped FLT phantom (8.0 cm−1 scattering) was loaded with 0 to 100 pmol Cy7 diluted in PBS with 2% Imeron 400 (Bracco Imaging, Milan, Italy) and 10% Lipovenös MCT 20% infusion solution (Fresenius Kabi, Bad Homburg, Germany). A calibration curve was calculated using linear regression analysis.

Pipet tips were used as inserts to be implanted into the dead mice (n = 3). They were filled with 60 pmol Cy7 in 5 μL of a PBS solution containing 2% Imeron 400 (Bracco Imaging, Milan, Italy) and 10% Lipovenös MCT 20% infusion solution (Fresenius Kabi, Bad Homburg, Germany). The fluorescent pipet tip inserts were closed with a lighter and placed in the esophagus, rectum, right kidney, and under the skin at the left flank. To exclude that differences in the fluorescence of the inserts would lead to misinterpretation of the data, the correct load of the inserts was confirmed in an FLT mouse phantom before placing them inside the dead mice.

The pipet tips in the dead animals were localized and segmented using CT or MRI scans. The insert segmentation was extended by 20 voxels to also capture the blurred signal.27

In Vivo Study on Immunoglobulin Biodistribution Immunoglobulin Purification and Labeling

Immunoglobulin G (IgG, monomer) and IgA (dimer) were purified from the serum of polymeric immunoglobulin receptor (pIgR)–deficient mice in which dimeric IgA accumulates in the blood. Sera of 15 pIgR-deficient mice were pooled to collect 3 mL. For IgG purification, the serum was loaded onto an ÄKTA start system (Cytiva Lifesciences, Marlborough, MA) using a 1-mL HiTrap Protein G High Performance column. IgG was eluted from the column using 0.2 mM glycin buffer at pH 2.5. The IgG-depleted serum was fractioned to purify IgA by size exclusion chromatography with an ÄKTA pure system (Cytiva Lifesciences, Marlborough, MA) and a Superdex 200 column. The purification yielded more than 1 mg for both isotypes. The pure antibodies were labeled according to the manufacturer's protocol with Alexa Fluor 750 (AF750) NHS Ester (ThermoFisher Scientific, Waltham, MA). The unreacted dye and the reaction buffer were removed by dialysis of the labeling reaction mixture against PBS using a Spectra/Por dialysis membrane (Carl Roth, Karlsruhe, Germany).

Animal Experiments

Eight female Crl:SKH1-Hrhr mice (Charles River Laboratories, Wilmington, MA) of 6–8 weeks were injected intravenous (IV) with a dye-equivalent of 2 nmol AF750-labeled immunoglobulin IgG or a dye-equivalent of 1.2 nmol IgA (4 mice per group) and the spasmolytic butylscopolamine (0.4 mg/kg body weight; PANPHARMA, Luitré, France) to reduce intestinal motility. Animal experiments were conducted in accordance with German legal requirements and approved by the local authority LANUV (Landesamt für Natur, Umwelt und Verbraucherschutz) Nordrhein-Westfalen.

Imaging

Mice were scanned under isoflurane inhalation anesthesia at 4.5 hours after immunoglobulin injection with CT-FLT (MILabs B.V.) and subsequently with MRI inside the FLT mouse holder with the parameters specified before. For scanning living mice, a respiration-gated MRI scan was performed with 500 milliseconds long gates resulting in an in vivo scan time of approximately 18 minutes.

Biodistribution Analysis

Normalization to the total body fluorescence directly after injection was performed as described before.1 Immunoglobulin concentrations are expressed as percent injected dose (ID)/cm3. For quantification, the respective organs were segmented (Figs. 2B, C). Please note that in MRI the fast-moving heart and lungs could not be separated, so they were segmented as a single unit. For reasons of comparability, the heart and lungs were also considered as 1 unit in the CT-aided analysis.

Statistical Analysis

Statistical analyses were performed using the GraphPad Prism 9 software (GraphPad Software, San Diego, CA). The segmented organ volumes of living mice in CT and MRI were compared via paired t tests (n = 8). To test for differences between the reconstructions in the insert quantification study and the biodistribution study, 2-way analysis of variance (ANOVA) was applied (n = 3 for the fluorescence inserts with Tukey multiple testing correction, n = 4 for each immunoglobulin group with Dunnett multiple comparison test). Two-way ANOVA was used to test for differences between quantifying the inserts, Tukey multiple comparison test was applied (n = 3). Data are expressed as mean ± SD. Differences were considered statistically significant for P < 0.05.

RESULTS

Although white petrolatum undergoes a chemical shift like oil, our study showed that it is the ideal marker for CT-MRI fusion. Petrolatum is visible in MRI and CT. Unlike water, oil, gelatin, agar, and glycerol, petrolatum is solid and nondrying, so it does not require replacement. Placing petrolatum markers in the mouse holder frame enabled CT-MRI fusion.

Segmentation Agreement and Marker Overlap Analyses Confirm Successful CT-MRI Co-registration

To evaluate the quality of CT-MRI fusion, this study evaluated the agreement of segmentations between CT and MRI in a LEGO phantom, dead mice, and living mice, respectively. After correcting for gradient linearity distortion and chemical shift, good agreement was observed (Fig. 1C, Supplemental Digital Content, Figure S1A, https://links.lww.com/RLI/A885). The overall good to medium segmentation agreement in living and dead mice was considered sufficient to fuse the 2 modalities (Fig. 2, Supplemental Digital Content, Figure S1A, https://links.lww.com/RLI/A885), in particular, when taking into account the limited resolution (280 μm) of the FLT. These data are further supported by the co-registration accuracy based on the markers with residual scores below 0.5 mm (Supplemental Digital Content, Fig. S1C–D, https://links.lww.com/RLI/A885).

Various Fluorescence Reconstruction Approaches Considering Information From CT, MRI, or Both Lead to Accurate Results

After ensuring the reliable fusion and overlap of CT and MRI data, we tested the quantification of CT-FLT, MRI-FLT, and CT-MRI-FLT reconstructions and whether the FLT reconstruction could be improved by extending organ scattering information (the asterisk indicates the inclusion of the extended scattering map). For this purpose, fluorescence inserts containing 60 pmol Cy7 were placed at 4 locations (esophagus, subcutaneous, rectum, and kidney) in dead mice (Fig. 3A). We considered the FLT reconstructions to be accurate if the mean fluorescence values were within a range of ±25% deviation from the inserted 60 pmol (Fig. 3B, Supplemental Digital Content, Figure S2, https://links.lww.com/RLI/A885). The variance, expressed by the standard deviations, indicated the precision of the measurements.

FIGURE 3: Fluorescence tomography reconstruction and quantification in dead mice. Fluorescence inserts in the esophagus (green), rectum (blue), kidney (pink), and under the skin (brown) were segmented (A, left panel fluorescence overlay in Batlow W,28 right panel 3D segmentation). The dye concentration (60 pmol Cy7) in the fluorescence inserts was quantified using 5 different reconstruction approaches (B, n = 3). Extending organ scattering information for the FLT reconstruction is indicated with an asterisk. Two-way ANOVA tests for differences in fluorescence signal quantification, depending on the insert location (n.a. indicates not available, see explanation in the results section). Data are shown as mean ± SD. The dotted lines depict ±25% deviation of the inserted 60 pmol.

The oral insert concentrations were overestimated by all the reconstruction approaches. The precision was higher for CT-aided than MRI-aided reconstructions (CT-FLT*: 95.7 ± 8.6 pmol, CT-FLT: 95.4 ± 8.7 pmol, CT-MRI-FLT*: 85.0 ± 21.2 pmol, MRI-FLT*: 86.7 ± 29.9 pmol, and MRI-FLT: 89.2 ± 30.3 pmol).

In contrast to the oral inserts, the concentration of the rectal inserts was underestimated in CT-aided reconstructions (note that the rectal inserts could not be reconstructed and analyzed in MRI because the signal-to-noise ratio in this region was too low for 8-cm-long mice due to the limited coverage of the rat volume coil). Computed tomography–FLT* (30.6 ± 12.3 pmol) was more accurate but less precise than the state-of-the-art CT-FLT reconstruction (22.1 ± 5.3 pmol). The combined CT-MRI-FLT* reconstruction with more accurate liver and kidney segmentation resulted in comparable values to the CT-FLT reconstruction (22.1 ± 5.8 pmol).

The subcutaneous inserts were accurately estimated. The purely MRI-aided reconstruction approaches were more precise (MRI-FLT: 58.2 ± 9.8 pmol, MRI-FLT*: 52.3 ± 7.5 pmol) than the CT-aided or combined reconstruction approaches (CT-FLT: 71.3 ± 25.6 pmol, CT-FLT*: 66.5 ± 23.0 pmol, CT-MRI-FLT: 68.7 ± 25.6 pmol).

Similar to the subcutaneous inserts, the renal inserts were accurately estimated (CT-FLT: 58.9 ± 16.1 pmol, CT-FLT*: 45.9 ± 27.9 pmol, MRI-FLT: 56.3 ± 16.6 pmol). The highest precision was observed for the extended scattering MRI-FLT* (50.0 ± 12.9 pmol) and CT-MRI-FLT* (56.8 ± 15.7 pmol) reconstructions. Overall, we observed location-dependent effects, such as concentration overestimation of oral inserts, underestimation of rectal inserts, and accurate reflection of subcutaneous and renal inserts.

MRI-FLT and CT-MRI-FLT Reconstructions Improve Biodistribution Analyses of Immunoglobulins

For the biodistribution study, we chose smaller mice than for the fluorescence insert experiments to better address the coverage of the MRI coil. In these mice, the biodistribution of AF750-labeled IgG (monomer) and IgA (dimer) was longitudinally tracked (Fig. 4, Supplemental Digital Content, Figure S3, https://links.lww.com/RLI/A885). Immunoglobulin-related fluorescence signals could be attributed to most organs with CT and MRI. However, especially organs involved in the distribution, metabolism, and excretion were better accessible with MRI-aided analyses. These include the liver with the gallbladder, kidneys, and spleen. The segmented liver volumes of healthy mice varied less in MRI than in CT, indicating a more reproducible segmentation (Supplemental Digital Content, Fig. S1B, https://links.lww.com/RLI/A885). The same was true for the kidneys and spleen due to the better-defined soft tissue contrast in MRI. In particular the spleen was only partially segmented in CT images as it could not be clearly distinguished from the liver, pancreas, and small intestine. Furthermore, MRI enabled brain and thyroid analysis, which are both organs where drug accumulation might be unwanted or harmful.

FIGURE 4: Fluorescence tomography reconstruction and quantification in vivo. AF750-labeled IgG (monomer, A, bottom left) and IgA (dimer, B, bottom left) were injected intravenous to track their biodistribution. Representative 3D fluorescence overlays (in Batlow W)28 can be found on the left side of the panels. Immunoglobulin concentrations were assessed 4.5 hours after injection (n = 4 per group). Two-way ANOVA tests for differences between the 5 FLT reconstruction approaches (not available organs are indicated with n.a.). Data are presented as mean percentage of the injected dose/cm3 ± SD.

To answer the 3 questions from the introduction, we conclude the following:

The similarity of the reconstructions demonstrated the successful development whole-body MRI-FLT, which can be used without CT in the future (Figs. 1, 3, 4, and Supplemental Digital Content, S1, https://links.lww.com/RLI/A885). Considering both the insert quantification and the biodistribution experiments, the additional organ-specific scattering coefficients that were used did not improve the accuracy of the FLT reconstruction (Figs. 3, 4). CT-MRI-FLT covered the entire mouse body for reconstruction (CT, Fig. 3) and provided the most detailed biodistribution results (MRI, Fig. 2, Figure 4, Supplemental Digital Content, Figure S1, https://links.lww.com/RLI/A885). Therefore, it was considered to be the most robust and comprehensive approach and only immunoglobulin concentrations determined by this analysis are reported below (all other concentrations can be found in Supplementary Tables S1–S2, https://links.lww.com/RLI/A885, but are comparable).

The distribution of IV injected IgG and IgA was assessed after 4.5 hours. IgG concentrated mostly in the liver and gallbladder (21.3% ± 2.9% ID/cm3 and 24.2% ± 6.2% ID/cm3). It concentrated less in the colon and spleen (4.9% ± 2.2% ID/cm3, 5.4% ± 4.7% ID/cm3) and even lower in the brain and thyroid (0.4% ± 0.2% ID/cm3, 2.5% ± 0.3% ID/cm3).

IgA distributed similarly to the liver (21.9% ± 6.6% ID/cm3) and the spleen (5.8% ± 6.1% ID/cm3). The MRI-aided analysis revealed that the IgA mainly accumulated in the gallbladder (54.1% ± 27.6% ID/cm3); its accumulation was significantly higher than for IgG (P < 0.0001, Supplemental Digital Content, Fig. S3, https://links.lww.com/RLI/A885). This and the higher concentration in the colon (9.4 ± 7.3% ID/cm3) resulted from the biliary excretion route of IgA.29 Furthermore, the MRI-aided reconstruction and segmentation revealed a comparable brain concentration but lower thyroid concentration than for IgG (0.4% ± 0.4% ID/cm3 and 1.1% ± 0.5% ID/cm3, respectively). The here assessed immunoglobulin biodistribution data are consistent with reports in the literature29–31 and demonstrate that the trimodal MRI-CT-FLT approach is highly capable of capturing differences in the biodistribution of these antibody isotypes.

DISCUSSION

To reconstruct FLT from planar images, photon scattering and absorption are computed considering the optical parameters and anatomical information of the heterogeneous mouse body.10,32–34 When neglecting the influence of dye labeling,1,2 the sensitivity of CT-FLT is just an order of magnitude below that of nuclear imaging.9 We hypothesized that the diagnostic value and the accuracy of FLT, in particular for whole-body imaging of mice, can be further improved by developing a trimodal CT-MRI-FLT approach with extended scattering information. Furthermore, we anticipated that adding MRI would improve the validity of biodistribution studies.

When comparing CT-, MRI-, or CT-MRI-aided FLT reconstructions, we observed similar fluorescence quantification results and biodistribution assessments in vivo. The limited coil coverage seemed to play only a minor role here, as the rectum was not part of the biodistribution analysis. However, the dye content measured with the 5 employed reconstruction approaches (CT-FLT, CT-FLT*, MRI-FLT, MRI-FLT*, and CT-MRI-FLT*) was either underestimated or overestimated in some locations in the ex vivo quantification study. The underestimation of dye concentrations in rectal inserts is consistent with a previous study, although a linear correlation between dye concentration and fluorescence signal was reported.9 Future studies might investigate this phenomenon more systematically than at only 4 locations to understand and normalize for this factor.

In our study, dye concentrations in oral inserts were overestimated. Because of the small tissue volume and the high heterogeneity in the cervical region, the assumption of the diffuse light propagation might not be fully applicable. Furthermore, the laser light may directly transcend the lateral regions, altogether resulting in an overestimation of dye concentrations in oral inserts.

In FLT reconstruction, the light propagation assumption relies on scattering as well as absorption. Unfortunately, it is not possible to differentiate whether nontransmitted light is scattered or absorbed. In our approach, the absorption was estimated from the excitation light transmission, resulting in a slightly lower value than in a previously published study,10 possibly due to interdevice variability or different settings in the reconstruction software (Supplemental Digital Content, Fig. S4, https://links.lww.com/RLI/A885). Furthermore, the absorption maps in Gremse et al10 relied on bilateral transmission scans, which may have compensated some scattering events, which is hardly possible with our current FLT system due to the connection between mouse holder and the system. In general, hemoglobin absorption in the near-infrared range is lower than in the UV range,35,36 but it varies between dead and living animals due to decreased absorption with increased blood oxygenation and clotting,35,37,38 and similarly, the optical properties of excised blood-drained organs change due to the loss of blood.

We believe that the assignment of inaccurate scattering coefficients is the most important reason for quantification errors. To obtain optical scattering coefficients for FLT, coefficients from literature were assigned.25 However, accurate scattering coefficients, which are determined among other factors by the cellular structure and shape,39 are difficult to obtain, and those reported in the literature highly vary between studies.25 Scattering is most commonly measured on excised tissues supported by simulations to obtain scattering coefficients using various methods such as reflectance confocal microscopy,40 optical coherence tomography,41 quantitative phase imaging,42 or integrated sphere measurements.43,44 However, these measurements are strongly affected by tissue shrinkage, loss, or coagulation of blood, all being factors that highly influence optical properties. Furthermore, the accuracy of these methods depends on the sample slice thickness, leading to variations in scattering events and diffuse light propagation.44 The accurate determination of scattering coefficients, and thus the improvement of scattering maps, is therefore crucial to improve quantification in optical imaging. An alternative to FLT for biodistribution studies, which is less susceptible to optical scattering events due to less acoustic scattering than optical scattering of tissue,45 would be photoacoustic tomography.46,47 It even allows label-free imaging with, for example, biomolecules48; however, quantification sensitivity of photoacoustic tomography is lower than that of FLT.49 Furthermore, photoacoustic tomography can suffer from ultrasound artifacts, which also make acoustic tomography difficult.50

Finally, we evaluated the influence of the different reconstruction schemes and the hybridization with CT, MRI, or both on the assessment of biodistribution with labeled immunoglobulins. Small changes in the detection of immunoglobulins mainly in liver, spleen, and kidneys were observed as MRI enabled a more detailed organ-segmentation than CT. For example, only with MRI-aided analyses that a reliable assessment of immunoglobulin accumulation in the spleen was possible, which, in line with the literature, was slightly higher for the more polymerized IgA.51 In addition, by segmenting the gallbladder in MRI scans, this study tracked the secretion of the polymeric immunoglobulins from the liver into the bile, which are transported via the pIgR to protect the mucosal epithelium.29 In contrast, the gallbladder-to-liver ratio of the monomeric IgG was not high. As we know that IgG is not excreted by the liver, the signals obtained are most probably related to fluorescence blurring, which affects the gallbladder due to its small size (approximately 2 mm in diameter).27 Reducing the size of the fluorescence measurement grid in this region by adding more data points might help to address this issue if very accurate values are required.

However, increasing the number of laser points would increase the imaging time and make it more difficult to capture dynamic biodistributions, especially in the early phases with blood half-lives far below 1 hour.1 In the future, it might be possible to develop “real-time FLT.” With adaptations to the current scanning protocol, it would be possible to perform FLT scans every 15 minutes by doing first the anatomical scan(s) and then inject the fluorescent probe via an IV catheter. Another or additional approach would be iterative reconstructions.52 There, the grid is not always scanned completely, but certain points are omitted. For the reconstruction of the scans, the values of these points are iterated. The need for such an approach depends on the kinetics of the observed probe. This is an interesting perspective for the future, but beyond the scope of this study. For the immunoglobulins, a later phase of biodistribution was studied, so the influence of this aspect is very limited.

The immunoglobulin biodistribution data might also be useful for future therapy studies. None of the immunoglobulins was highly concentrated in the thyroid and the brain, which was expected, pointing to the accuracy of our approach. However, correct fluorescence quantification in these organs will be of high importance for other drugs and biomolecules as they may represent either the target tissue or a risk area for (long-term) drug toxicity.

To achieve similarly detailed fluorescence reconstruction results with CT, contrast agents could be used to improve soft tissue contrast. However, these may compromise animal welfare due to additional IV injection or the induction of adverse effects. The latter have been reported for iodinated contrast agents due to their accumulation in kidneys and the thyroid, which may also change the biodistribution of the probe of interest.12 Furthermore, when doing both native and contrast-enhanced CT scans, the tumor physiology may get disturbed, which has been reported previously even with low-dose CT scans.13

Although the biodistribution assessment was more differentiated with MRI, it was difficult to assign fluorescence spots in off-center locations of the mouse, despite a rat volume coil being applied. Therefore, the fluorescence in rectal inserts could not be reconstructed and analyzed with MRI and also dye concentrations determined for oral inserts were less precise compared with CT. Scanning the entire mouse body can be achieved by using larger MRI coils, such as rabbit coils, but this would reduce resolution and signal-to-noise ratio,53 and thus the image quality. If this is not feasible, a more pragmatic approach would be to scan the mouse twice and reconstruct the images adjacent to each other, but it would increase the acquisition and postprocessing time. For central or subcutaneous locations, the lack of the mouse head and tail regions in the MRI scans did not strongly affect FLT reconstruction and quantification. In line with Graves et al,33 the dye concentration in subcutaneous regions was quantified accurately, which is particularly valuable for biodistribution studies with subcutaneous tumors.1,2,10,54 Magnetic resonance imaging–aided reconstructions and analyses quantified dye concentrations at subcutaneous locations even more precisely, probably due to the more accurate delineation of the skin.

Altogether, considering the advantages and disadvantages of CT-aided (higher coverage) and MRI-aided (detailed organ biodistribution analysis) optical reconstructions, a trimodal approach will lead to the most comprehensive and accurate results. Besides using MRI for FLT reconstruction, adding MRI allows to additionally perform a range of functional analyses for characterizing tissue properties to assess therapeutic or toxic effects.14–17,55

In summary, the comprehensiveness and robustness of biodistribution studies can be improved by a CT-MRI-FLT hybrid imaging approach. The required hybridization was realized via markers placed in the mouse holder, which is a pragmatic, adaptable, and generalizable approach for combining multiple imaging modalities.

ACKNOWLEDGMENT

The authors thank the Institute for Laboratory Animal Science Aachen for providing dead mice.

REFERENCES 1. Schraven S, Rosenhain S, Brueck R, et al. Dye labeling for optical imaging biases drug carriers' biodistribution and tumor uptake. Nanomedicine. 2023;48:102650. 2. Hage C, Gremse F, Griessinger CM, et al. Comparison of the accuracy of FMT/CT and PET/MRI for the assessment of antibody biodistribution in squamous cell carcinoma xenografts. J Nucl Med. 2018;59:44–50. 3. Biancacci I, Sun Q, Möckel D, et al. Optical imaging of the whole-body to cellular biodistribution of clinical-stage PEG-b-pHPMA-based core-crosslinked polymeric micelles. J Control Release. 2020;328:805–816. 4. Batti Angulski AB, Cohen H, Kim M, et al. Molecular homing and retention of muscle membrane stabilizing copolymers by non-invasive optical imaging in vivo. Mol Ther Methods Clin Dev. 2023;28:162–176. 5. Rakhilin N, Yang B, Spilker ME, et al. Volumetric imaging of optically cleared and fluorescently labeled animal tissue (VIOLA) for quantifying the 3D biodistribution of nanoparticles at cellular resolution in tumor tissue. J Control Release. 2023;354:244–259. 6. Deng L, Chen J, Li Y, et al. Cryo-fluorescence micro-optical sectioning tomography for volumetric imaging of various whole organs with subcellular resolution. iScience. 2022;25:104805. 7. Foo W, Wiede A, Bierwirth S, et al. Automated multicolor mesoscopic imaging for the 3-dimensional reconstruction of fluorescent biomarker distribution in large tissue specimens. Biomed Opt Express. 2022;13:3723–3742. 8. Stokes C, White EF, Toddes S, et al. Whole CNS 3D cryo-fluorescence tomography shows CSF clearance along nasal lymphatics, spinal nerves, and lumbar/sacral lymph nodes. J Imaging. 2023;9:45. 9. Rosenhain S, Al Rawashdeh W, Kiessling F, et al. Sensitivity and accuracy of hybrid fluorescence-mediated tomography in deep tissue regions. J Biophotonics. 2017;10:1208–1216. 10. Gremse F, Theek B, Kunjachan S, et al. Absorption reconstruction improves biodistribution assessment of fluorescent nanoprobes using hybrid fluorescence-mediated tomography. Theranostics. 2014;4:960–971. 11. Kunjachan S, Gremse F, Theek B, et al. Noninvasive optical imaging of nanomedicine biodistribution. ACS Nano. 2013;7:252–262. 12. Di Gregorio E, Arena F, Gianolio E, et al. The interaction between iodinated x-ray contrast agents and macrocyclic GBCAs provides a signal enhancement in T1-weighted MR images: insights into the renal excretion pathways of Gd-HPDO3A and iodixanol in healthy mice. Magn Reson Med. 2022;88:357–364. 13. Baier J, Rix A, Darguzyte M, et al. Repeated contrast-enhanced micro-CT examinations decrease animal welfare and influence tumor physiology. Invest Radiol. 2023;58:327–336. 14. Minhas AS, Sharkey J, Randtke EA, et al. Measuring kidney perfusion, pH, and renal clearance consecutively using MRI and multispectral optoacoustic tomography. Mol Imaging Biol. 2020;22:494–503. 15. Jerome NP, Boult JK, Orton MR, et al. Modulation of renal oxygenation and perfusion in rat kidney monitored by quantitative diffusion and blood oxygen level dependent magnetic resonance imaging on a clinical 1.5T platform. BMC Nephrol. 2016;17:142. 16. Gerwing M, Hoffmann E, Kronenberg K, et al. Multiparametric MRI enables for differentiation of different degrees of malignancy in two murine models of breast cancer. Front Oncol. 2022;12:1000036. 17. Meissner A, Garcia-Serrano AM, Vanherle L, et al. Alterations to cerebral perfusion, metabolite profiles, and neuronal morphology in the hippocampus and cortex of male and female mice during chronic exposure to a high-salt diet. Int J Mol Sci. 2022;24:300. 18. Ren W, Li L, Zhang J, et al. Non-invasive visualization of amyloid-beta deposits in Alzheimer amyloidosis mice using magnetic resonance imaging and fluorescence molecular tomography. Biomed Opt Express. 2022;13:3809–3822. 19. Stuker F, Baltes C, Dikaiou K, et al. Hybrid small animal imaging system combining magnetic resonance imaging with fluorescence tomography using single photon avalanche diode detectors. IEEE Trans Med Imaging. 2011;30:1265–1273. 20. Zhang Y, Zhang B, Luo J, et al. In vivo tomographic imaging with fluorescence and MRI using tumor-targeted dual-labeled nanoparticles. Int J Nanotechnol Nanomed. 2013;9:33. 21. Bley TA, Wieben O, François CJ, et al. Fat and water magnetic resonance imaging. J Magn Reson Imaging. 2010;31:4–18. 22. Gremse F, Stärk M, Ehling J, et al. Imalytics preclinical: interactive analysis of biomedical volume data.