Phenotyping placental oxygenation in Lgals1 deficient mice using 19F MRI

Placental hypoperfusion and hypoxia are key drivers in complications during fetal development such as fetal growth restriction and preeclampsia. In order to study the mechanisms of disease in mouse models, the development of quantitative biomarkers of placental hypoxia is a prerequisite. The goal of this exploratory study was to establish a technique to noninvasively characterize placental partial pressure of oxygen (PO2) in vivo in the Lgals1 (lectin, galactoside-binding, soluble, 1) deficient mouse model of preeclampsia using fluorine magnetic resonance imaging. We hypothesized a decrease in placental oxygenation in knockout mice. Wildtype and knockout animals received fluorescently labeled perfluoro-5-crown-15-ether nanoemulsion i.v. on day E14-15 during pregnancy. Placental PO2 was assessed via calibrated 19F MRI saturation recovery T1 mapping. A gas challenge with varying levels of oxygen in breathing air (30%, 60% and 100% O2) was used to validate that changes in oxygenation can be detected in freely breathing, anesthetized animals. At the end of the experiment, fluorophore-coupled lectin was injected i.v. to label the vasculature for histology. Differences in PO2 between breathing conditions and genotype were statistically analyzed with linear mixed-effects modeling. As expected, a significant increase in PO2 with increasing oxygen in breathing air was found. PO2 in Lgals1 knockout animals was decreased but this effect was only present at 30% oxygen in breathing air, not at 60% and 100%. Histological examinations showed crossing of the perfluorocarbon nanoemulsion to the fetal blood pool but the dominating contribution of 19F MR signal is estimated at > 70% from maternal plasma based on volume fraction measurements of previous studies. These results show for the first time that 19F MRI can characterize oxygenation in mouse models of placental malfunction.

oxygenation 14 . A major advantage of small animal MRI is the possibility to follow dynamic processes over time in repetitive sessions, which is not possible in histological examinations.
The partial pressure of oxygen (PO 2 ) is the most relevant parameter for the body's oxygen sensing systems 15 and all 1 H MR methods (BOLD/TOLD/perfusion MRI) provide only surrogate markers of PO 2 . More direct techniques such as polarographic electrodes, fiberoptic probes or optical imaging suffer from limited penetration depth and are invasive for deeper tissue 16 . A particularly promising technique for preclinical studies of placental blood oxygenation saturation (SO 2 ) is optoacoustics but it cannot directly assess PO 2 17-19 . Therefore, there is a strong need for other spatially-resolved techniques to measure the absolute partial pressure of oxygen of the placenta in vivo. Since the fluorine ( 19 F) longitudinal relaxation rate R 1 = 1/T 1 of several perfluorocarbons (PFCs) correlates with PO 2 , 19 F MRI has been used to study deep tissue oxygenation in animal models of disease including tumors, kidney injury or cerebral hypoperfusion [20][21][22][23][24][25] . The investigation of dynamic vascular changes and placental perfusion and oxygenation by MRI is a promising opportunity to explore angiogenesis-related pregnancy disease.
Pregnancy acts as a cardiac stress model inducing in some cases adverse cardiac events in healthy women without any previously known cardiovascular disease [26][27][28] . As a consequence of such cardiac stress 5-7% of all pregnancies develop preeclampsia which is associated with hypertension after the 20th week of gestation (with or without proteinuria), in conjunction with fetal growth restriction, maternal endothelial dysfunction and chronic immune inflammation [29][30][31] . Preeclampsia imposes a maternal increased risk of cardiovascular disease death later in life, independent of other measured risk factors. Moreover, preeclampsia is a major underlying cause of late fetal and early neonatal mortality 29,30 . A poor utero-placenta circulation secondary to inadequate remodeling compromises nutrition and oxygenation of the fetus and is associated with fetal growth restriction 29,30 .
Among the immunoregulatory and angiogenic factors, galectin-1 (gal-1), a member of a family of carbohydrate-binding proteins, has been shown to modulate several processes associated with placentation, promotion of maternal tolerance toward fetal antigens and regulation of decidual vascular expansion during the pre-placentation period [32][33][34] . Blocking gal-1-mediated angiogenesis with anginex, a 33-mer cytokine-like artificial β-peptide, results in a spontaneous preeclampsia-like syndrome in mice, mainly by deregulating processes associated with good placentation and maternal spiral artery remodeling 35 . Consistent with these findings, using gal-1 deficient dams we demonstrated the development of a preeclamptic-like phenotype in which mice developed gestational hypertension, proteinuria, smaller litters and progressive glomerulosclerosis. In addition to this, gal-1 deficient mice also demonstrated endothelial dysfunction, abnormal maternal decidual arteries, increased vascular resistance in the uterine arteries and poor placental development 35 .
The goal of this exploratory study was to investigate the potential of 19 F MRI to measure placental oxygenation after i.v. injection of a Perfluoro-15-crown-5-ether (PFCE) emulsion in the gal-1 deficient Lgals1 −/− knockout mouse, hypothesizing a PO 2 deficit in this model of preeclampsia.

Results
Method validation by gas challenge. To Fig. 1. Figure 2 shows typical volumes of interest overlaid on anatomical 1 H images along with T 1 fits of mean 19 F signal-to-noise ratio (SNR) and the final result for mean PO 2 in WT and KO mice at varying oxygen levels in breathing air. The shape of the fitted curves varied substantially, which indicated heterogeneity in PO 2 between animals, between placentas within one animal and between different breathing conditions. Descriptive statistics showed an expected increase in SNR on T 1 -weighted images (Fig. 3A), a decrease in T 1 (Fig. 3C) and increase in PO 2 with increasing oxygen in breathing air, while relative 19 F concentration S 0 remained similar. Mixed-effects modeling of SNR(TR i , i = 1, 2, 3, 4), T 1 and PO 2 between genotype and percent oxygen in breathing air are summarized in Table 1 and confirmed this observation. A significant effect of percent oxygen in breathing air on SNR, T 1 and PO 2 was found, even for SNR at the least T 1 -weighting (TR = 5000 ms) indicating an expected strong effect of oxygen level on PO 2 in the placenta. The fitted equilibrium signal S 0 = SNR(TR >> T 1 ) is a marker of local contrast agent concentration and was neither different between animals nor between breathing conditions. www.nature.com/scientificreports/ 19 F MRI of placental PO 2 in preeclampsia mouse model. No effect of genotype was found in any of the parameters in the full statistical linear mixed-effects model but a significant interaction between percent oxygen and genotype at 30% O 2 in breathing air (Table 1). In line with this, descriptive statistics did not show major genotype differences at higher oxygen levels in breathing air (Fig. 3). We further investigated genotype differences at 30% oxygen by linear mixed-effects modeling ( Table 2). A significant effect of genotype was found for PO 2 , T 1 and the T 1 -weighted images SNR(TR 1 = 318 ms) and SNR(TR 2 = 719 ms) whereas this effect vanished for less T 1 -weighting at longer TRs. In Lgals1-knockout mice PO 2 was lower compared to wildtypes (KO: 38 ± 52 mmHg, WT: 102 ± 35 mmHg, data expressed as estimated mean (ME) ± 95% confidence interval (CI), p = 0.016) i.e. T 1 was higher (KO: 2.18 ± 0.43 s, WT: 1.66 ± 0.30 s, ME ± 95%CI, p = 0.020). Accordingly, SNR was lower in transgenic animals for T 1 -weighted images at lowest TR (TR = 318 ms, KO: 1.26 ± 0.14, WT: 1.45 ± 0.10, p = 0.008) and second-lowest TR (TR = 719 ms, KO: 2.12 ± 0.33, WT: 2.57 ± 0.23, ME ± 95%CI, p = 0.009).
Histology. In order to qualitatively assess the distribution of 19 F agent in the maternal and fetal blood pool and in cells of the placenta, fluorescent microscopy of the maternal blood vessels (fluorophore-coupled lectin, green), the 19 F agent (Rhodamine, red) and cell nuclei (DAPI, blue) was performed. Figure 4, shows the 19 F agent labeled with Rhodamine crossed the placental barrier and penetrated the maternal and fetal circulation within the labyrinth area of the placenta.

Discussion
So far, most studies on oxygenation 19 F MRI have focused on tumors and liver. We hypothesized that the placenta is a particularly favorable organ to study with this technique since a large volume fraction consists of maternal plasma which can be reached via simple i.v. injection. In a three compartment model (maternal, fetal and trophoblast pool), a recent diffusion MRI study estimated the maternal blood volume fraction of the placenta at 64.4% (fetal pool: 23.7%, trophoblast pool: 11.9%) and optical imaging confirmed this (~ 70%:30% ratio of maternal:fetal blood pool volume) 13 . In our qualitative histological examinations, we found most fluorescence signal originating from maternal blood but some signal was also present in both trophoblast and fetal pool. Although dissociation of the fluorophore from PFCE nanoemulsion particles is possible 36 , the vesicular patterns of Rhodamine signal hints toward true phenomenon of particles crossing the placental barrier. Mechanisms that regulate the transfer of nanoparticles across the placenta and fetal circulation include simple difussion, active transport, phagocytosis and endocytosis 37,38 . Considering the average PFCE nanoparticle size of 100-200 nm, phagocytosis/endocytosis seem the most likely route. This hypothesis is supported by the finding that caboxylate-modified polystyrene beads with diameters between 20 and 500 nm injected intravenously in pregnant mice accumulated in the placenta via trophoblast uptake 39 . Taken our histological observations together with the quantitative argument on volume fractions, we assume that at least 70% of the contribution to PO 2 in our study originates from maternal plasma. The contribution is probably higher since diffusivity in the fetal blood pool is orders of magnitude higher than in the other two compartments 13 which further diminishes the fetal 19 F MR signal as summarized in Fig. 5. Since PO 2 is likely to be different for each of the three compartments, 19 F MRI signal fractions could be estimated via equilibrium signals S0 maternal , S0 trophoblast , S0 fetal from multiexponential fitting. Adding the assumption of compartmental volume fractions from diffusion imaging could dissect the exact amount of PFCE crossing. Although the 19 F MR sensitivity in our study was not sufficient to perform such fine-grained models with 6 fitting parameters (S 0 and T 1 per compartment), future gains in SNR will enable these types of analyses. www.nature.com/scientificreports/ We assumed a steady state of PFCE concentration in the different compartments of the mouse placenta, which is a valid assumption since S 0 , a marker of local 19 F concentration, was not different between breathing conditions over a period of ~ 1.5 h. During this time, internalization of PFCE nanoparticles by circulating monocytes/ macrophages is very likely 40 . There is an oxygen gradient between mitochondria and plasma, which may lead to apparently lower PO2 values of intracellular vs. extracellular PFCE. Using oxygen-sensitive microscopy of green fluorescence protein, intracellular oxygen gradients were found to be very small in monolayer cell culture (~ 0.03 mmHg/µm) 41 and we thus did not correct for this source of bias.
Placental PO 2 under normal and pathological conditions. The most relevant outcome parameter in pathology of placental circulation is the oxygenation of the fetus. We found lower placental pO 2 in Lgals1 knockout dams. Due to the low statistical power in our experiments, this finding needs to be confirmed in a hypothesis-driven study. Another limitation of the current study is that most PO 2 signal originates from maternal pool, i.e. it provides only an indirect marker of fetal oxygenation. A recent optoacoustic study, however, has shown that changes in SO 2 in the feeding maternal artery, the placenta and the fetus are directly correlated during a gas challenge 17 . The same phenomenon is expected for PO 2 in the range of 0-128 mmHg when PO 2 and SO 2 are tightly linked through the oxygen dissociation curve (OCD). Under 30% O 2 , the PO 2 values found in this study in wildtypes (~ 100 mmHg) are higher than those known from optoacoustics under normoxia with 20% oxygen (SO 2 ~ 50%, i.e. PO 2 ~ 30 mmHg using the OCD). The differences cannot be attributed to the higher level of oxygen in breathing air alone. Inaccuracies in the in vitro calibration of R 1 and PO 2 , discrepancies in anesthesia depth, use of NO 2 in our study and the inability of optoacoustics to capture dissolved oxygen in blood present further obstacles in direct comparison of results. For hyperoxia (100% O 2 ) we show dramatic increases in partial pressure of oxygen (PO 2 > 200 mmHg) which can only be explained by a significant amount of dis-  19 F MRI parameters in Lgals1 wildtype (WT, black circles) and knockout (KO, gray circles) placentas at 30%, 60% and 100% oxygen in breathing air. Shown are (A) signal to noise ratio on one of the T 1 -weighted images, (B) fitted relative 19 F concentration S 0 , (C) fitted T 1 relaxation time and (D) calculated PO 2 ~ 1/T 1 . Increasing the amount of oxygen in breathing air led to an apparent increase in SNR, decrease in T 1 and increase in PO 2 but also an increase in variance. Parameters indicative of decreased oxygenation in knockout animals were only obvious at 30% oxygen. Red lines show mean, red and blue areas correspond to 95% confidence intervals and standard deviation, respectively. www.nature.com/scientificreports/ solved oxygen. This observation is in line with other studies of maternal arterial PO 2 in sow, mare and ewe 42 .
In addition, besides the phenomenon of dissolved oxygen in blood, PFCs themselves are able to transport gases and this effect becomes stronger when breathing high levels of oxygen, a phenomenon which is well described in PFC-based therapy of hypoxia 43,44 and could open the field of placental theranostics.  25,36 . The experiments of this study were performed at a field strength of 7 T using a double-tuned room temperature 40 mm transmit/receive birdcage resonator coil and the robust turbo spin echo saturation recovery pulse sequence. Within reasonable 20 min imaging time, we managed to acquire sufficient SNR for T1 fitting, albeit at the border of detectability for the highest T 1 -weighting. The SNR was lower compared to in vitro experiments and this could be explained by motion which is omnipresent when imaging the abdomen in vivo. Consequently, our study has several limitations. First, voxel-wise analysis of PO 2 was not possible due to error propagation in the calculation of PO 2 from T 1 . Second, smoothing was necessary to improve SNR at loss of resolution. Third, only few T 1 -weighted images were acquired within reasonable time introducing error on the two-parameter fit. These sources of variance and PO 2 -unrelated changes of relaxivity  www.nature.com/scientificreports/ e.g. due to local field distortions at tissue/air interfaces in the bowel, can account for the physiologically impossible values of PO 2 below 0 or larger 760 mmHg (1 atmosphere) in our data. Assuming a near-linear relationship of SNR efficiency with field strength (sample-dominant noise), an improvement in sensitivity by factor > 2 is to be expected for the currently strongest MRI system at 21.1 T 46 . Our first results with cryogenic coils for mouse brain 19 F MRI have shown a similar increase in sensitivity of factor ~ 2 to 3 compared to room temperature coils 21,47 . A transfer of this technology to abdominal imaging in pregnancy is pending but feasible through adaptation of coil geometry. More SNR efficient T 1 mapping such as Look-Locker methods are generally more prone to artefacts but should further be explored for abdominal 19 F MRI 48 . Finally, the 19 F MR signal is limited to maternal blood pool and other, spatially sparse regions of the mother. This fact can be used to accelerate imaging using compressed sensing reconstruction. First studies show that a decrease of imaging time at identical SNR by factor > 2.5 is possible for point-like sources of signal 49,50 . Absence of genotype differences were observed for the two higher levels of oxygen in breathing air, so even in the current study design, a significant improvement in SNR efficiency by factor 3 reduction in scan time can be anticipated when only interested in genotyping. Taken all of these efforts together, an extraordinary gain in SNR efficiency up to an order of magnitude seems possible with modified study designs and technology available already now or in the near future.
Technical considerations: 19    www.nature.com/scientificreports/ lives or complex response of T 1 to PO 2 (Perfluorodecalin). Based on a recent comparison of PFCs for preclinical research 51 , we decided to use Perfluoro-15-crown-5-ether since this compound has a strong, narrow single resonance from 20 magnetically equivalent 19 F nuclei per molecule and a very long half life in blood, allowing long imaging times in a biologically stable state. For oxygenation imaging, the response of PFCE relaxation rate R 1 to changes in PO 2 follows a simple linear relationship and the slope is in the medium/upper range compared to other compounds 52 . However, the extremely long half live in the liver 51 limits the use of this compound to preclinical animal studies. In addition, it needs to be determined if the 19 F agent has any undesired effect on fetuses. Generally, PFCs are biologically inert and very well tolerated by the body even in high doses. For adults, safety of high doses of i.v. injected PFC emulsions similar to the one used in this study, have been confirmed in clinical studies. Future improvements in SNR, pulse sequences for complex spectra 53 and shortening of imaging times will enable the use of clinically more favorable PFCs such as PFOB, which has a similarly strong 1/T 1 to PO 2 response for the CF 3 group (unpublished in vitro data). Further improvement include the use of higher concentrated emulsions in order to increase SNR and PEGylation of particles to prolong systemic circulation time 54 .

Conclusion
Combining advances in small animal MRI hardware and 19 F agent synthesis, this study presents the first important step in using 19 F MRI for placental phenotyping. Detection of decreased PO 2 in Lgals1deficient mice highlights the potential of the technique in a mouse model of preeclampsia, one of the most detrimental pregnancy complications. We foresee two important use cases of placental oxygenation 19 F MRI. First, it presents a screening method in animal models of placental malfunction, especially when oxygenation is a primary outcome measure and is uncoupled from surrogate markers such as perfusion or blood oxygenation SO 2 . Second, since 19 F MRI provides absolute values of PO 2 , it bears potential to calibrate less invasive techniques based on conventional 1 H measurements such as BOLD/TOLD MRI 55 . This is the first study to characterize placental oxygenation in vivo during uneventful pregnancy and a pathological disorder such as preeclampsia using 19 F MRI.

Methods
Animals. Inbred 129/P3J lectin, galactoside-binding, soluble, 1 (Lgals1) wildtype (WT) and deficient (KO) mice were maintained in our animal facility with a 12L/12D cycle 35 . Eight-to 10-week-old virgin female Lgals1 WT or KO mice were mated with 8-to 14-week-old Lgals1 WT or KO males respectively. Females were inspected daily for vaginal plugs; sighting a vaginal plug was designated as day 0 of pregnancy. Pregnant Lgals1 WT (normal pregnancy, n m = 3 mothers/n p = 19 placentas) or KO (preeclampsia model, n m = 4/n p = 17) female mice were subjected to 19  www.nature.com/scientificreports/ factor = 2, ΔTE = TE = 14.3 ms, BW = 10 kHz, NA = 5, TA = 19:50 min). A narrow excitation and refocusing pulse bandwidth (1 kHz) was chosen to avoid chemical shift artefacts from isoflurane signal. In order to transfer anatomical volumes of interest (VOI) from the anatomical 1 H scan to the 19 F T 1 weighted images, the center and orientation of both slice stacks were identical. Each time after changing breathing gas O 2 levels, we allowed > 10 min until the center of k-space of the 19 F scans in order to ensure an equilibrium of blood gases including a 5 min resting period and the 5-7 min during 1 H MRI acquisition (Fig. 1).
MRI PO 2 analysis. MRI analysis was performed blinded to the genotype of animals. To reduce the impact of noise, T 1 weighted images were preprocessed in ImageJ (v1.52a, https ://image j.nih.gov) using the following steps: 1. Voxel-wise conversion of signal intensity into SNR. In order to mitigate impact of non-Gaussian noise distribution in low SNR magnitude images, we used the approximation by Gudbjartsson 57 and transferred to the preprocessed 19 F SNR maps. Mean 19 F SNR was measured for each TR and exported into MATLAB (version R2014b, MathWorks, Natick, USA). The relaxation time T 1 of the placenta was calculated by fitting the saturation recovery signal model The fitted equilibrium signal S 0 is proportional to the 19 F concentration and was analyzed separately. PO 2 was calculated using the in vitro calibration of a previous study 21 Histology. At the end of the MRI acquisition, pregnant mice received 100 µL DyLight 488 Labeled Lyco-persiconEsculentum (Tomato) Lectin (LEL, TL, Vector Laboratories, BiozolDiagnostica) dissolved in 100 µL PBS i.v. over 5 min. Animals were euthanized by cervical dislocation. The entire pregnant uterine horns were dissected and first rinsed in 0.1 M PBS, cryoprotected in Tissue-Tek (VWR), frozen and kept at − 80 °C until processing. Serial cryosections from whole implantations on E14/E15 were cut at 8 μm. The slides were washed in TBS for 5 min and nuclei in all sections were counterstained by incubating 5 min in DAPI solution, followed by washing and mounting in ProLong Gold (Invitrogen, Thermo Scientific; 99-904-02). Sections were analyzed using a Pannoramic Digital Slide Scanners MIDI microscope (3DHistech).

Statistics.
Mean SNR(TR i ) (i = 1, 2, 3, 4), i.e. one value for each of the four preprocessed 19 FT 1 -weighted images, T 1 relaxation time, PO 2 and S 0 in the placenta were statistically evaluated in SPSS (v. 25.0., IBM Corp., Armonk/NY, USA). Mothers were assumed to be the statistically independent unit, i.e. the analysis of placentas was corrected for nesting. Linear mixed-effects modeling (SPSS MIXED) was used with genotype as the between factor and percent oxygen in breathing air as a within factor. Fixed effects of genotype, percent oxygen and their interaction were analyzed. Significant interaction effects were further investigated for an effect of genotype using post-hoc linear mixed model for each level of percent oxygen. Significant effects were further investigated using unpaired two-tailed t-tests. Each parameter of interest was treated independently without post-hoc correction between parameters.

Data availability
Work instructions, MRI raw data, processed data used for calculating PO 2 and detailed statistical output can be found on https ://doi.org/10.5281/zenod o.38762 71. www.nature.com/scientificreports/