Non-invasive imaging platform reveals a potential tumourigenicity hazard of systemically administered cells

The number of clinical trials using cell-based therapies is increasing, as is the range of cell types being tested, but without a thorough understanding of cell fate and safety. Therefore, there is a pressing need for monitoring of cell fate in preclinical studies to identify potential hazards that might arise in patients. Utilising a unique imaging toolkit combining bioluminescence, optoacoustic and magnetic resonance imaging modalities, we assessed the safety of different cell types by following their biodistribution and persistence in mice. Our imaging studies suggest that the intra-arterial route is more hazardous than intravenous administration. Longitudinal imaging analysis over four weeks revealed that the potential of mouse mesenchymal stem/stromal cells (mMSCs) to form tumours, depended on administration route and mouse strain. Clinically tested human umbilical cord (hUC)-derived MSCs formed growths in 15% of animals that persisted for up to three weeks, indicating a potential tumourigenicity hazard that warrants further testing.


Introduction
Cell-based regenerative medicine therapies (RMTs) have the potential to treat various diseases 1 , but the risk of tumour formation is a primary safety concern 2 . Mesenchymal stem/stromal cells (MSCs) isolated from bone marrow, adipose tissue or umbilical cord are being tested in clinical trials, but in many cases, preclinical safety data are not available. Bone marrow-derived MSCs have been used for many years and appear safe 3 , but a review of adipose-derived MSCs concluded that while adverse events are rare, they nevertheless do occur, and are likely to be related to underlying health conditions of the patients or administration route 4 . Human umbilical cordderived (hUC)-MSCs have only recently been introduced in clinical trials, with more than 50% of these initiated within the last 3 years (Supplementary Table 1). hUC-MSCs are less immunogenic than other types of MSCs, which contributes to their attraction as clinical RMTs. However, because of their low immunogenicity in combination with higher proliferative behaviour these cells may 4 also pose a greater potential risk 5 , yet until now, their safety profile has not been robustly assessed. The importance of preclinical safety testing is highlighted by a recent report where a tumour developed in a patient's spinal cord following intrathecal administration of stem cells 6 .
The most common way to administer cells systemically in small animals is via the intravenous (IV) route 7 , delivering cells directly to the lungs where they are sequestered as a consequence of the pulmonary first-pass effect [8][9][10][11][12][13] . Although the IV route is also frequently used in clinical trials, administration via the arterial circulation is not uncommon. For instance, clinical trials testing the potential of cell therapies to treat myocardial infarction administer cells into the coronary arteries or left cardiac ventricle 14,15 , while in patients with peripheral artery disease or stroke, intra-arterial injection via the femoral or carotid artery, respectively, is frequently employed 16 . Intra-arterial administration will also lead to systemic distribution to other organs, including the brain, and cells passing through the blood-brain barrier could pose an important safety concern. However, a detailed analysis of cell fate after intra-arterial cell administration has so far not been reported 17 .
Non-invasive imaging technologies have opened up exciting new possibilities for preclinical assessment of the safety of cell therapies by monitoring cell biodistribution and persistence through longitudinal in vivo cell tracking; however, a platform approach using multimodal imaging for the safety assessment of cells has not been previously implemented. Preclinical imaging technologies for cell tracking, some of which have clinical relevance, include magnetic resonance imaging (MRI) to detect cells labelled with superparamagnetic iron oxide nanoparticles (SPIONs), multispectral optoacoustic tomography (MSOT) to detect cells labelled with gold nanorods or near-infrared red fluorescent protein (iRFP) [18][19][20][21][22][23] , and bioluminescence imaging (BLI) for the detection of cells expressing the genetic reporter, firefly luciferase [24][25][26] . Genetic reporters are particularly advantageous because signals are only generated from living cells, thus allowing the monitoring of cell proliferation and tumour growth, and avoiding problems based on nanoparticle dissociation from cells, which can lead to false positive signals. However, the spatial resolution of 5 BLI is poor, making it difficult to precisely locate the cells 24 . By contrast, both preclinical MSOT and MRI have much higher spatial resolution (150 µm and 50 µm, respectively), providing details of the inter-and intra-organ distribution of administered cells. Moreover, as MRI is routinely used in the clinic, it provides a bridge for preclinical and clinical studies.
Here, we have implemented a multi-modal imaging approach comprising BLI, MSOT and MRI, to assess biodistribution and fate of different cell types following venous and arterial administration in healthy mice. Some of these cell types are currently being used in clinical trials, including hUC- Table 1), kidney-derived cells 27 and macrophages 28 . We show that in a small number of mice, hUC-MSCs started to proliferate over time. Although these hUC-MSC growths eventually regressed, our data raise safety concerns regarding the use of these cells in clinical trials.

Results
Whole body biodistribution of different cell types following intravenous (IV) and intracardiac (IC) administration Bioluminescence imaging showed that IV delivery of ZsGreen + /Luc + mMSCs, mKSCs, hKCs and hUC-MSCs, resulted in signals exclusively in the lungs, while signals from IV-administered macrophages were also located more posteriorly (Fig. 1a). This was expected because macrophages are known to traverse the lungs and populate other organs, such as the liver and spleen. In contrast, intraarterial delivery via the left ventricle (from now on referred to as intra-cardiac (IC)) resulted in a whole-body distribution of all cell types (Fig. 1a).
Organ-specific ex vivo imaging within 1h of mKSCs being administered IV confirmed the signal was limited to the lungs (Fig. 1b, d). In contrast, after IC administration, bioluminescent signals were detected in the brain, heart, lungs, kidney, spleen, and liver (Fig. 1b, d). A comparable ex vivo biodistribution was observed for mMSCs and hKCs following IV and IC administration (not shown), 6 while with hUC-MSCs, we found low but detectable signals in other organs besides the lungs following IV administration ( Supplementary Fig. 1). IV-administered macrophages were found predominantly within the lungs by ex vivo imaging (Fig. 1c), but weaker signals were also detected in the spleen and liver ( Supplementary Fig. 2), confirming the in vivo signal distribution. Ex vivo analysis of macrophages after IC injection showed signals in all organs (Fig. 1c, e). (a) BLI immediately after administration, showing that cells were always confined within the lungs after intravenous (IV) administration, but distributed throughout the body after intracardiac (IC) administration; an exception were the macrophages which showed also a more posterior signal after IV administration. Luminescence intensity scale has been adjusted individually for each cell type and is described in Supplementary Table 2. Ex vivo bioluminescence imaging of organs within 5h of administration of (b) mKSCs or (c) macrophages confirmed the in vivo cell biodistribution.
Organs are indicated as kidneys (k), spleen (s), liver (li), lungs (lu), heart (h) or brain (b). Quantification of the bioluminescence signal intensity of organs ex vivo post (d) mKSC or (e) macrophage administration. Values represent the mean signal intensity measured in each organ and normalised to the total flux from all organs (n = 3 each group). Error bars represent standard error. (f) Mean pixel intensity of GNR-labelled macrophages measured via multispectral optoacoustic tomography for a period of 5 hours post IV administration, displaying the kinetics of their accumulation in the spleen and liver. Arrow indicates the time point at which the cells were administered.
To monitor the temporal dynamics of macrophage migration, cells were labelled with GNRs, injected IV, and monitored continuously for 4.5h using MSOT. Signal intensity began to increase immediately in both the liver and spleen until around 90 min (Fig. 1f), but remained close to basal levels in the kidney, consistent with BLI ex vivo analysis (Fig. 1e, f). However, when GNR-labelled macrophages were administered IC, increases in signal intensity in the kidney were comparable to those in the liver and spleen 4h post-administration ( Supplementary Fig. 3c).
Cell distribution within organs using high-resolution magnetic resonance imaging Since the spatial resolution of BLI is poor, we used MRI to evaluate the intra-organ biodistribution of ZsGreen + /Luc + /SPION + mMSCs after IV or IC administration, focussing particularly on the brain and kidneys. Following IC injection, T2 * weighted imaging revealed hypointense areas distributed homogenously throughout the brain (Fig. 2a), and localised in the cortex of the kidneys (Fig. 2b).
However, hypointense contrast was not detected in the brain or kidneys of IV-injected mice, confirming that IV-administration does not deliver mMSCs to either of these organs (Fig. 2a, b).
Post mortem MR imaging of extracted organs performed at higher resolution confirmed the hypointense contrast throughout the brain and in the renal cortex of IC-injected mice (Fig. 2a, b). Histological analysis of ZsGreen expression by fluorescence microscopy in combination with Prussian Blue staining of SPIONs showed that labelled cells were located in the glomeruli (Fig. 2c).
ZsGreen and Prussian Blue signals corresponded to the same spatial location, indicating that hypointense contrast in vivo was unlikely to result from false-positive detection of SPIONs (e.g. released from dead cells). To determine whether IC-administered cells had undergone extravasation, we performed confocal imaging of IB4-stained blood vessels. This demonstrated that ZsGreen + mMSCs were physically trapped in the lumen of microcapillaries (Fig. 2d), suggesting that the cells did not cross the blood brain barrier or the glomerular filtration barrier.

Short-term fate of IC-injected cells
To determine how long the cells persisted in major organs we injected 10 6 ZsGreen + /Luc + /SPION + mMSCs into the left ventricle of BALB/c mice and tracked their fate in vivo by MRI and BLI, and post mortem by MRI and fluorescence microscopy ( Fig. 3a). On the day of injection, whole-body distribution of IC-administered mMSCs by bioluminescence signals was observed, while in the kidneys, MRI revealed hypointense contrast specifically in the cortex. By 24h, bioluminescence signal intensity decreased, suggesting cell death. Correspondingly, fewer hypointense areas were observed in the renal cortex by MRI, supporting the disappearance of SPION-labelled cells. By 48h, bioluminescence was no longer detectable in the abdominal region, nor was any significant hypointense SPION contrast observed in the kidneys with MRI. This was confirmed by highresolution MRI of organs ex vivo, showing a decrease in contrast in the renal cortex over time, and a decrease in the frequency of ZsGreen + mMSCs in kidney glomeruli by fluorescence microscopy (Fig. 3a). Changes in the T2* relaxation time in the renal cortex indicated the relative number of SPION-labelled cells present at each time point. T2* was significantly lower on the day of cell administration ( Fig. 3b) than at baseline but then increased towards baseline levels at 24h and 48h.
Because the liver is the major organ for clearance of blood-transported particulates, we quantified 10 the hepatic T2* relaxation time, which revealed a subtle but significant decrease from baseline through to 48h (Fig. 3c). These results suggest that following cell death, SPIONs accumulate predominantly in the liver and are not retained by the kidneys. where green fluorescence corresponds to ZsGreen expression and blue fluorescence to DAPI staining. Arrowheads indicate individual glomeruli. Scale bar corresponds to 100 µm. T2 * relaxation time of (b) kidney cortices or (c) liver before (baseline) and up to 2 days after cell administration. The T2 * relaxation time in the cortex of the kidney was significantly lower on the day of cell administration (day 0, mean = 7.98 ms +/-SE = 0.29) than at baseline (14.56 +/-0.32 ms; One-way ANOVA, p < 0.001). The T2 * relaxation time then increased towards baseline levels at day 1 (12.57 +/-0.50 ms) and day 2 (13.19 +/-0.23 ms), and by day 2 the difference compared with baseline levels was no longer statistically significant. In the livers, T2* relaxation time revealed a subtle but significant decrease in relaxation time from baseline through to day 2 (baseline, 7.19 +/-0.29 ms; day 0, 5.48 +/-0.38 ms; day 1, 5.10 +/-0.16 ms; day 2, 5.02 +/-0.94 ms; One-way ANOVA, p = 0.006). See Supplementary Table 3 for Tukey pairwise comparisons.
Effect of administration route on the long-term biodistribution and fate of mMSCs To assess the effect of administration route on the long-term fate of cells, ZsGreen + /Luc + mMSCs were administered to BALB/c SCID mice by IC or IV routes, and biodistribution monitored by BLI at multiple time points for 28 days. While both IC and IV injection resulted in the typical immediate biodistribution patterns by 24h (Fig. 1a), by 96h following IV and IC administration, the bioluminescence signal was undetectable, indicating loss of cells via cell death (Fig. 4a). Continued imaging over time showed that bioluminescence signals began to increase again in animals after IC injection from around day 14, consistent with tumour development, but not in animals after IV injection. The increase in signal was particularly prominent in the hindquarters of all five ICinjected mice at day 14, and increased further until day 28 (Fig. 4a, Supplementary Fig. 4a).
Detailed analysis of animals after IV administration of mMSCs revealed bioluminescence signals in the lungs of one mouse increased over time (Supplementary Fig. 4b). Overall, whole-body bioluminescence intensity initially decreased following both IC and IV administration, and subsequently increased rapidly in the IC-injected mice (Fig. 4b-d). Osteosarcoma formation after IC administration of mMSCs Multiple abnormal growths were present in IC-injected BALB/c SCID mice, predominantly in skeletal muscle surrounding the femurs, but also in muscle near the hips, ribs, and spine (Fig. 5a, f), suggesting tumours had formed. Tumour sites corresponded to foci of intense BL signals which could also be identified using T2 weighted MR imaging (Fig. 4e). Furthermore, T2 weighted MR imaging allowed us to detect an abnormal mass in the lungs of one (out of three) IV-injected mouse that displayed an intense bioluminescence signal (Fig. 4e, Supplementary Fig. 4b). Although cells of the mMSC line have been suggested to home to the bone marrow 29 , flow cytometry analysis showed the bone marrow was negative for ZsGreen + cells ( Supplementary Fig. 5).
Histologically, tumours were characterised by atypical solid proliferation of spindle cells associated with multifocal formation of pale amorphous eosinophilic material (osteoid). The tumours were therefore classified as osteosarcomas (Fig. 5h, j, k). Frozen sections of the tumour tissue exhibited specific ZsGreen fluorescence (Fig. 5i), further confirming the neoplasms originated from mMSCs.

Formation of mMSC-derived tumours in different mouse strains
To determine whether tumours developed because the BALB/c SCID mice were immunocompromised, we investigated the long-term fate of the mMSCs following IC administration in three different immunocompetent mouse strains: BALB/c (same genetic background as mMSCs), FVB (unrelated inbred strain), and MF1 (unrelated outbred strain). The biodistribution immediately after injection was similar between the strains, but at day 28, only the BALB/c mice displayed bioluminescence signals as high as those in the BALB/c SCID mice (Fig. 5b-e).
Moreover, the timing and location of tumour formation was consistent in all immunocompetent and immunocompromised BALB/c mice. In the FVB and MF1 strains, mMSC foci tended to form in similar locations as with the BALB/c mice, but bioluminescence signals were weaker. Although 14 signal intensity gradually increased in FVB mice from d7 to d28, in MF1 outbred mice, signals increased initially up to d21, but then started to decrease as the mMSC foci began to regress.  Table 1). When following the fate of IV-and ICadministered hUC-MSCs in BALB/c SCID mice, we found that in most cases, BLI signals became weaker within a few days of administration, and remained undetectable for the duration of the study (8 weeks) (Fig. 6a, b). However, in a small number of mice (~15%) UC-MSC foci had developed in locations outside the major organs (Fig. 6c, red arrows). Although these foci initially expanded, they then appeared to regress, and by d21 were barely detectable, and did not reappear during the remaining 5 weeks of the experiment. Representative BLI of mice administered with 10 6 hUC-MSC via the IC or IV route. The signal was progressively lost shortly after administration, with no evidence of malignant growth. (b) Mean whole body quantification of the bioluminescence signal up to day 28 as obtained with two different cell doses. Error bars represent SE. (c) BLI images from mice that displayed signal that persisted up to or beyond day 7 (dose: 10 6 cells, ventral orientation). In all cases, the signals had disappeared by day 21 and not returned by day 56. Images in blue frames (a, c) are presented in a lower intensity scale (1.0 x10 4 -1.0 x10 5 p/s/cm 2 /sr) to display weaker signals.

Discussion
Here, we have employed a novel platform approach of non-invasive preclinical imaging encompassing BLI, MRI and MSOT to assess the biodistribution and persistence of a range of mouse and human cell types following IV and IC administration in healthy mice. These cells included mouse MSCs, kidney stem cells and macrophages, as well as human kidney-derived cells and hUC-MSCs, the latter being already tested as cell therapies in clinical trials. As expected, immediate analysis after IV administration revealed that apart from macrophages, all other cell types were mostly sequestered in the lungs, although small numbers of hUC-MSCs could be detected in other organs following ex vivo analysis. After IC administration, all cell types showed a widespread distribution. However, irrespective of the administration route, analysis using all three imaging technologies determined that cells disappeared from major organs within 24-48 hours, which based on the loss of BLI signals, was likely due to cell death. The observation that cells are cleared very quickly from the major organs following IC administration indicates that the arterial route poses no significant advantage for cell therapy administration. By contrast, our long-term tracking analysis over four to eight weeks, provides the first evidence that arterial administration of cells may carry a higher tumourigenicity hazard. Therefore, this striking finding suggests that IV administration of cell therapies is safer for clinical applications. We thus recommend that further investigations are urgently needed before conducting clinical trials where cells are delivered arterially.
Our platform of imaging techniques was also able to provide some mechanistic insight into the fate of cells after administration. Macrophages have been previously shown to home to the liver and spleen after passage through the lungs 30 . However, the dynamics of this homing process had not been described. Using multi-modal BLI and MSOT, we could monitor macrophage accumulation in the liver and spleen for 4.5h continuously at high spatial resolution. We found that labelled macrophages immediately started to accumulate in liver and spleen, particularly in the first ~90 min, which indicated that some of the macrophages instantly passed through the pulmonary circulation.
While BLI has the advantage of highly sensitive body-wide detection of luciferase-expressing cells, its spatial resolution is poor, which prevents organ-focussed imaging. To visualise cells within major organs such as kidney and brain, and monitor their fate over time, we implemented a bimodal approach comprising BLI and MRI, taking advantage of the high spatial resolution of MRI in addition to the high sensitivity of BLI and the fact that luciferase activity is dependent on cell viability [20][21][22]24 . Detailed analysis of the biodistribution of mMSCs after IC injection using in vivo, and subsequently ex vivo MR imaging techniques revealed that SPION-labelled cells were scattered throughout the brain, while in the kidneys, they were restricted to the cortical regions.

Ex vivo histological staining and fluorescence microscopy demonstrated that cells in the kidneys
were found only within the glomeruli, bounded by endothelial cells within the microvasculature.
Similarly, cells in the brain were only localised within the microvasculature, indicating that they lack the capacity to pass through the blood brain barrier. These results demonstrate that the mMSCs cannot extravasate into the brain and kidneys, and are in line with our observation that tumours were not found in these organs after four weeks.
Surprisingly, during long-term cell tracking of the BALB/c-derived mMSCs, we observed tumour formation in skeletal muscle following IC administration to a similar degree in immune-competent BALB/c mice as in BALB/c SCIDs. mMSCs also gave rise to tumours in an unrelated inbred strain, albeit at a slower rate, while in an unrelated outbred strain, small foci of mMSCs expanded at early time points and later regressed. Taken together, these data suggest that the adaptive immune system might not be able to recognise tumours derived from syngeneic MSCs (equivalent to autologous MSCs in human applications), and that the genetic background of the host appears to have an effect on the propensity of MSCs to form tumours. This could be a concern for human trials using autologous MSCs where the ability of the cells to form tumours may not be detected by the recipient's immune system. Furthermore, the results suggest that the risk of tumour formation might depend on undefined genetic factors that would vary from patient to patient.
Our observation that mMSCs distributed to most organs following IC injection, but tumours were predominantly localised in the skeletal muscles and not within the organs they originally appeared in, raises the question of how tumour formation is regulated in different organs and tissues. Our data indicate that the cells had a 'survival advantage' in muscular tissue, but not in the brain and the kidneys, from which they failed to extravasate. We hypothesise that following IC administration, a small number of MSCs were able to extravasate from the capillaries in the skeletal muscle where they started to proliferate. The mechanisms that regulate the ability of the mMSCs to extravasate and form tumours in the skeletal muscle but not in other organs are not known, and further analysis is required to determine the molecular and cellular factors controlling this process. Our results also show that the cells failed to home to and populate the bone marrow, which is surprising given the cells had been originally isolated from the bone marrow 31 . The D1 mMSC line used here has not previously been reported to generate invasive tumours, since subcutaneously injected cells provided no evidence of metastasing, even if they proliferated at the injection site 32,33 . Our observation that the mMSCs did not form tumours outside the lung following IV administration is therefore consistent with this finding.
Since these observations suggested that arterial administration of MSC-based cell therapies could have important safety implications, we followed the fate of hUC-MSCs, which are currently being used in several clinical trials (Supplementary Table 1). While in most animals the cells became undetectable within a few days after IV administration, in a few mice the cells persisted longer, albeit transiently, in other body regions where their presence was not expected. We suggest that this unusual behaviour is not linked to cell size, because the hUC-MSCs were not smaller than mKSCs or mMSCs, but could possibly be due to their surface proteins, allowing some of the cells to escape the lungs 11,34 . The observation that hUC-MSC foci appeared in a small number of mice, grew in size, but later disappeared, was difficult to explain, especially given that the mice were SCIDs and thus lacked an adaptive immune system. It is possible that the cells eventually elicited a xenogeneic response involving macrophages and natural killer cells 35 , after initially suppressing the native immune system, which is one of their central properties 36,37 . Alternatively, the hUC-MSCs may have expanded in the animal but then become senescent and died, irrespective of the host's ability to mount an immune response. Thus, after an 8-week period of cell tracking, we could not Average cell diameter was estimated by measuring the volume of a cell pellet in a packed cell volume (PCV) tube according to the manufacturer's instructions (Techno Plastic Products, Switzerland). The cell diameter was calculated using the formula: where V corresponds to the pellet volume, and c to the number of cells in the pellet.
For MR tracking, cells were labelled with diethylaminoethyl-dextran coated SPIONs synthesised in house as previously described 42,43 Table 6).  At the respective study end points, mice were culled and organs with any visibly identifiable tumours imaged ex vivo by BLI. Kidneys were cut coronally for ex vivo imaging, and all other organs were imaged whole. Bioluminescence signals of whole live mice or individual organs ex vivo were 25 quantified by drawing regions of interest (ROIs) from which the total flux (photons/second) was obtained. The relative signal intensity from each organ was calculated as a percentage of the signal intensity from all organs. For ex vivo kidney imaging, the ROI was drawn around all four kidney halves and a single value for total bioluminescence signal was recorded.
Images were recorded at the following wavelengths: every 10 nm from 660 nm and 760nm, and every 20 nm from 780 nm and 900 nm, at a rate of 10 frames per second and averaging 10 consecutive frames. All mice were allowed to adjust to the imaging system for 15 minutes prior to recording data. For monitoring biodistribution of macrophages after IV administration, a 15 mm section of the abdomen to include the liver, kidneys and spleen of the mice was imaged repeatedly for a total of 4.5 hours; 30 minutes into the imaging the BALB/c mice (n = 3) received 10 7 macrophages via a tail vein catheter. For the IC imaging a 15 mm section of the abdomen was imaged once, followed by an ultrasound (Prospect 2.0, S-Sharp, Taipei city) guided injection of 10 7 macrophages into the left ventricle of the heart of 3 BALB/c mice. Mice were then returned to the photoacoustic imaging system for imaging as previously described. Data was reconstructed and multispectral processing was performed to resolve signals in the liver, kidney and spleen for GNRs.
Regions of interest were drawn around the liver, right kidney and spleen ( Supplementary Fig. 3) to generate mean pixel intensity data.

MR imaging
ZsGreen + /Luc + /SPION + mMSCs (10 6 ) were administered to BALB/c mice IV (n = 2) or IC (n = 2 for short-term analysis; n = 5 for longitudinal tracking). The biodistribution of cells in the brain and kidney was imaged with a Bruker Avance III spectrometer interfaced to a 9.4T magnet system

Statistical Analyses
Statistical analyses were performed using Minitab 17 statistical software. A one-way ANOVA (analysis of variance) was used to compare multiple groups. When an ANOVA resulted in a statistically significant result (p < 0.05), a Tukey pairwise comparison was performed in order to determine which groups were significantly different. The Tukey pairwise comparison assigned each group at least one letter, and groups that did not share a letter were significantly different from one another.