Quantitative MRI of Gd‐DOTA Accumulation in the Mouse Brain After Intraperitoneal Administration: Validation by Mass Spectrometry

In mice, intraperitoneal (ip) contrast agent (CA) administration is convenient for mapping microvascular parameters over a long‐time window. However, continuous quantitative MRI of CA accumulation in brain over hours is still missing.

4][5][6] Vascular involvement has been established in animal models of Alzheimer's disease, 7 epilepsy, 8 depression, 9 cerebral malaria, 10 amyotrophic lateral sclerosis, 11 and multiple sclerosis. 12Robust and noninvasive quantification of microvascular parameters, such as the cerebral blood volume (CBV), cerebral blood flow, and altered endothelial permeability is required for the early detection, characterization, and therapeutic follow-up of such diseases.Although MRI methods such as arterial spin labeling and diffusion weighted approaches exploiting endogenous contrast mechanisms are under development, 13,14 most MRI approaches such as dynamic contrast enhanced (DCE), dynamic susceptibility contrast and steady state contrast enhanced techniques still rely on intravenously (iv) administered contrast agents (CA).The efflux of metabolic waste products from the interstitial fluid involves perivascular and perineural pathways constituting the glymphatic system, with waste ending up in the subarachnoid space and draining into meningeal and nasal lymphatic vessels and cervical lymph nodes. 15Like the vascular system, the glymphatic system can be impaired in neurodegenerative diseases, including Alzheimer's disease and cerebral small vessel disease. 16Recently, DCE-like MRI protocols based on intrathecal or intracisternal CA administration have been proposed to study the pathways of the glymphatic system in rodents. 17,18These protocols require the quantification of the CA in tissue from T 1 -weighted dynamic MRI, 19 which is not straightforward because it requires acquisition of a proton density (PD) or T 1 weighted acquisition before CA administration for reference, a quasi-linear or otherwise known relationship between the signal and CA concentration over a wide range, minimization of confounding effects from flow and T 2 (*) changes as well as whole brain coverage with a reasonable time resolution.Alternatively, pre-and postcontrast T 1 relaxometry can be used but is generally more time consuming.
A time resolved MRI approach has been proposed to monitor CA uptake into the brain parenchyma upon intravenous 20 and intraperitoneal (ip) administration. 21So far, this approach has only been used to quantify the regional CBV and CBV changes upon a pharmacological challenge, 21 similarly to iron oxide particle-based pharmacological MRI approaches. 22However, similarly to DCE-MRI, it allows to detect accumulation of the CA beyond the vascular compartment. 23Such CA accumulation can either be triggered by a drug or a functional change or because of a pathology affecting the endothelial permeability 20,23 or the efficacy of the glymphatic efflux.This approach does not yield quantitative parameters for the blood flow and the endothelial permeability.However, its easy implementation is of interest in mouse models characterized by a slow accumulation of the CA and/or an impaired clearance since the regional CA uptake rate is appropriate to characterize the tissue, disease induced changes, or pharmacological effects.In many applications, the aim is to detect if and when the CA accumulates in the brain tissue.
Against this background, we aimed to validate this time resolved MRI approach for the quantitative assessment of the CA concentration, and to investigate the spatiotemporal distribution in mouse brain upon its ip administration.We also employed this approach to detect if the CA accumulates in brain tissue upon pharmacological challenge with ip hyperosmolar mannitol.

Materials and Methods
This pilot study was carried out in accordance with French regulations (decree 2013-118) and the European directive (2010/63/EU) on the protection of animals used for scientific purposes after approval by our institutional committee on ethics and the ministry of higher education and research (project authorization #31714).

Animals and Experimental Procedure
Female 14-to 16-week-old C57Bl/6JRj mice (N = 25; Janvier Labs, Le Genest-Saint-Isle, France), weighing 21.6 AE 1.6 g and housed in groups in an enriched specific pathogen-free environment with food and water ad libitum, were randomly allocated to one of two groups (group 1: control N = 10; group 2: mannitol N = 15) by a zootechnician unaware of the study design.Further blinding was impractical since the researchers were involved in all experimental and analysis procedures.The study was carried out during the light phase of the 12-hour light/dark cycle.Catheterization of the peritoneal cavity with two 24G catheters and MRI acquisitions were performed under isoflurane (1.5%-2%/air; Vetflurane, Virbac, Centravet, Lapalisse, France).The mouse eyes were lubricated (Ocry-gel, Tvm laboratory, Lempdes, France).In the magnet, rectal temperature (35-37 C) and respiration (80-120 minutes À1 ) were monitored continuously.
In 20 mice (10 from group 1 and 10 mannitol-treated mice from group 2), the CA was removed from the blood pool by transcardial perfusion (Supplementary Materials and Methods).The unperfused mice (5 mannitol-treated mice from group 2) were euthanized by cervical dislocation, the brains were extracted, freezeclamped in liquid nitrogen, and stored at À80 C. The delay between the last MRI acquisition and the interruption of the blood flow by cervical dislocation or perfusion was approximately 2 and 10 minutes, respectively.
The mice were positioned prone.All images were acquired in the axial plane and covered the entire brain from the olfactory bulbs to the cerebellum, except for single-slice T 1 maps, which were acquired at bregma level.Field homogeneity was accomplished using a B 0 map shim technique.
Before Gd-DOTA administration, the imaging protocol (Table 1; Fig. 1) consisted of a brief angiography, and in a subset of 8 mice of five T 1 -weighted images to compute T 1 maps.The MRI technique used to map the CBV and the Gd-DOTA uptake required an acquisition at thermodynamic equilibrium (termed PD) to normalize the time resolved T 1 -weighted signal (trMRI), which was acquired with an inversion recovery-prepared fast gradient echo with experimentally optimized parameters (Table 1) to suppress the signal from the brain parenchyma and the blood prior to Gd-DOTA injection. 21A baseline signal was acquired over 5 minutes before Gd-DOTA injection.The trMRI signal was then acquired over 135 minutes interrupted once at 120 minutes by a second angiography to confirm the absence of motion by superposition onto the precontrast angiography and by a T 1 -weighted highresolution image for anatomical reference.

Mass Spectroscopy
The frozen brain samples were desiccated and prepared for gadolinium quantification by inductively coupled plasma mass spectrometry (ICP-MS model 7800, Agilent Technologies, Les Ulis, France).Methodological details are provided in the Supplementary Materials and Methods).The gadolinium concentration was converted from μgÁg À1 dry tissue to millimolar (mM) using a dilution factor of 1/5 upon desiccation and a brain density of 1046 g L À1 . 25
Based on simplifying assumptions generally made for DCE-MRI (and detailed in Supplementary Materials and Methods), the trMRI signal S(t) can be analyzed using Eq. 2 (Table 2). 20Eq. 2 represents the distribution volume fraction (DVf) as long as the CA is confined to a tissue compartment at a concentration above C crit = 5 mM.At this concentration, the short T 1 of this compartment allows its signal to reach the thermodynamic equilibrium while the signal from compartments with long T 1 is negligible.The tissue signal is therefore proportional to the volume fraction of the distribution compartment.A constant trMRI signal shortly after CA arrival in the brain tissue is known to arise from the vascular compartment. 20,21ontrary to previous studies aiming at the quantification of the DVf of the CA, 20,21,23 we did not assume that the C crit was attained or maintained over the entire observation time.Instead, the normalized trMRI signal was converted into longitudinal relaxation rate R 1 26 (Fig. S1a), from which regional CA concentration was estimated (Eqs. 3 and 4, Table 2).The normalized trMRI signal is almost linearly dependent on CA concentration for 1.18 < R 1 < 10 second À1 (or 850 > T 1 > 100 msec), corresponding to tissue concentrations of 0.16 < C < 2.7 mM (Fig. S1a).
A vascular concentration > C crit is necessary to quantify the cerebral blood volume fraction (BVf), yet the signal is saturated leading to an underestimation of the CA concentration in tissue while it is confined to the vascular space (Fig. S1b).The trMRI signal might decrease again, reflecting a CA concentration decrease below C crit .However, any observed trMRI signal increase arises from CA accumulation within the extravascular compartment (Fig. S1b), 23 even if the vascular CA concentration does not remain above C crit over the entire observation interval.
The conversion depends on the tissue relaxivity of Gd-DOTA assumed to be in the interval 2.8 mM À1 s À1 ≤ r 1 ≤ 3.6 mM À1 s À1 for whole blood 27 and plasma 28 at 7-T and 37 C and on the baseline relaxation rate R 10 (= T 10

À1
) of brain tissue.To avoid lengthy acquisitions of T 1 maps for each mouse, Gd-DOTA concentration maps were computed using an average R 10 of 0.6 second À1 .Concentration maps were generated by averaging 20 trMRI repetitions up to 15 minutes and up to 135 minutes after Gd-DOTA administration, when available, otherwise up to the last trMRI time point.The brain tissue was delineated manually (by ATPB, 19 years of experience in rodent models) on each slice of the PD acquisition Eq. 1 in Ref. 21 Longitudinal relaxation rate R 1 Eq. 7 in Ref. 26 Tissue concentration  (Fig. S2), taking care to only include structures collected for ICP-MS.Extracerebral cerebrospinal fluid (CSF), cranial nerves, and the pituitary gland were therefore excluded.The Gd-DOTA concentration was then averaged within the delineated brain mask (16480 AE 708 voxels).The ventricles were also segmented (by IV, 8 years of experience in brain segmentation, 407 AE 148 voxels).Maximum intensity projections were generated from the 3D MRI volumes after segmentation of the brain.

Statistical Analysis
Analysis was performed with Prism software (version 9.0; GraphPad Software, San Diego, CA, USA).The measured parameters were the MRI estimate of the Gd-DOTA concentration averaged over the entire brain at the end of the observation interval and the quantification of gadolinium by ICP-MS.Bland-Altman, one-tailed Spearman correlation, D'Agostino-Pearson test for normal distribution, and Pearson correlation analyses were performed.Values were expressed as mean AE standard deviation.A P-value <0.05 was considered statistically significant.

Results
The average T 10 for intraventricular CSF and brain tissue was 2.266 AE 0.222 seconds and 1.625 AE 0.039 seconds, respectively (N = 8), i.e.T 10 > 1 second as necessary for signal suppression prior to CA administration (Fig. 2).From the constant trMRI signal observed in brain tissue in the 10-15 minutes interval after CA administration an average BVf in the order of 0.02 was measured.
The Gd-DOTA concentration in brain tissue was in the order of 0.2 mM 15 minutes after CA administration.Two hours later, it reached 0.5 mM on average in brain tissue and above 1 mM in CSF.This CA concentration range falls within the fairly linear part of the signal to concentration dependence (Fig. S1a).With three exceptions, mannitol-treated mice showed signs of cardiorespiratory failure before the intended end of the trMRI acquisition, in which case the acquisition was interrupted, the animal euthanized, the brain processed as planned and data included in the analysis.Due to motion artifacts (N = 2), failed MRI acquisition (N = 1), and failed ICP-MS quantification (N = 1), both measures were only available for 16 perfused and 5 unperfused mice (Fig. 3).Our study showed no evidence of Gd-DOTA loss from the ventricular CSF due to transcardial perfusion or brain sampling (Fig. 3b).Despite a significant correlation (Fig. 3; Table S1), the MRI-derived concentration was noticeably higher since it included the CA in the blood pool.This is reflected by the bias of 0.35 mM in the Bland-Altman analysis (Fig. 3c).For unperfused mice, the two techniques yielded more comparable values, in particular for r 1 = 3.6 mM À1 s À1 .An r 1 of 2.8 mM À1 s À1 would result in MRI measures that are a factor of 3.6/2.8= 1.2857 higher.
Immediately after injection, Gd-DOTA was mainly seen in the vascular system (Fig. 4).Furthermore, Gd-DOTA accumulation occurred in ventricular CSF within the first hour and in periventricular parenchyma, particularly in the hypothalamus, beyond 1 hour after administration.The Gd-DOTA uptake was also predominant in the olfactory bulb.The spatiotemporal CA accumulation pattern was similar for all mice, though with variable degree of accumulation among mice (Figs. 3 and 4).Mannitol treatment, despite the relatively low dose, increased tissue uptake of Gd-DOTA in 11/12 mice.

Discussion
This study shows that, in addition to the quantification of the regional cerebral BVf, the proposed trMRI approach is able to detect CA accumulation over time and to monitor its uptake kinetics in a quantitative and direct way.The approach is suited for the detection of vasoactive or bloodbrain barrier modulating properties of pharmacological agents, and for the study of the spatiotemporal distribution of CAs, including efflux via the glymphatic pathways.
The measured BVf of 0.02 is in agreement with previous studies using the trMRI technique with an ip Gd-DOTA dose of 6 mmol/kg at 4.7-T 21 or other approaches, 29,30 confirming that the quantitative measure is independent of B 0 and CA dose.Knowledge of the tissue T 1 or the vascular input function is not required to map this parameter.Only the PD acquisition was required for quantification.
The T 10 map was only acquired for validation purposes.Due to the longer baseline T 10 , the CA concentration estimation in ventricular CSF has the highest inaccuracy when assuming a global R 10 of 0.6 second À1 .About 15 minutes after CA administration when the tissue concentration was still low, the concentration can be underestimated by up to 30% in CSF and overestimated by up to 15% in brain structures with lower T 10 such as white matter.Two hours after administration, the high tissue CA concentration reduced the relative error to 5%.
Although individual T 10 maps covering the entire brain may increase the accuracy of the quantification in particular for low CA concentrations, we believe that the use of a single representative T 10 value for brain tissue is acceptable for most preclinical studies allowing to save overall acquisition time or to lengthen the observation time of the trMRI acquisition to ensure the desired pharmacological effect can be observed.
Compared to steady-state techniques such as the CA-based vascular space occupancy technique that shares the same assumptions, the described trMRI technique has the advantages of more reliable blood nulling, and the detection of CA extravasation owing to the time resolved acquisition.This study showed that even slow and subtle CA accumulation in brain parenchyma after ip administration in control mice can be detected and quantified with the time resolved MRI approach.It can be used for pharmacological MRI when using a CBV-modifying drug or in response to hypercapnia, 26 or to detect disruption of the blood-brain barrier (BBB) by focused ultrasound, ischemia, intracarotid hyperosmolar mannitol administration, or death.
Specifically, ip administration routes are of interest and comparable with iv routes when using CAs such as Gd-DOTA and MnCl 2, 24,31 albeit with a slight delay resulting from the time required for the CA to reach the vasculature.In addition to the ease of (repeated) catheterization and the less stringent restrictions on injectable volume in mice, 32 ip administration resembles an iv continuous infusion due to the CA reservoir in the peritoneal cavity and lengthens the observation time window.
The ip administration route is also less invasive and less technically challenging than the intrathecal, intracisternal or intracerebroventricular administration of CA, and this study showed that it may result in a comparable spatial distribution in the brain within the same observation time.In this study, slow ip bolus injection led to a reasonable observation time.A dose of 10 mmol/kg Gd-DOTA (higher than the recommended dose of 6 mmol/kg 21 ) was administered in this exploratory study involving MRI at higher magnetic field and euthanasia for brain sampling.Alternatively, an ip continuous infusion protocol can be established following a lower initial dose.
Hyperosmolar mannitol is known to increase the BBB permeability but high doses also raise the cerebral BVf. 33oth changes lead to an increase of the tissue Gd-DOTA concentration and therefore to a trMRI signal increase.However, the correlation between in-vivo and ex-vivo measures in mannitol-treated and transcardially perfused mice would be compromised if BVf increases were the source of signal increase.The ICP-MS-based Gd quantification in perfused mice confirmed that gadolinium indeed reached the interstitial space.
The spatial distribution is evocative of the one observed in rodent studies with intracisternal or intraventricular CA or fluorescent tracer administration 17,18 highlighting the role of the glymphatic system in the transport of the CA to the brain parenchyma, in particular under hyperosmolar conditions. 34mitations Technical advances such as parallel imaging and sparse sampling and reconstruction techniques can improve the compromise between spatial and temporal resolution of the MRI acquisition. 35As other dynamic techniques, it has a high sensitivity to motion that is difficult to correct using image registration due to the signal nulling of brain and extracerebral tissue before CA arrival.
Time resolved MRI without pharmacokinetic modeling is not suited for the quantification of the microvascular permeability, but mapping the CA concentration over time allows estimation of a regional accumulation rate, which is generally expected to qualitatively reflect microvascular leakiness. 36However, as is the case for the volume transfer constant K trans , the contribution of permeability, blood flow and-with long observation times-interstitial diffusion cannot be disentangled.Despite dynamic monitoring of the MRI signal change, pharmacokinetic modeling yields a set of quantitative microvascular parameters that is not time-resolved and requires determination of the arterial input function and an appropriate compartment model, including the vascular, interstitial, and CSF space. 36This study aimed at validating a direct and functional MRI technique for the mapping of the accumulation rate over hours.
While the parenchymal CA concentration was the measure ultimately aimed for, the tissue concentration of the CA mapped by MRI is a surrogate measure and the combined result of the microvascular BVf, the plasma concentration of the CA, the microvascular leakiness, and the existence of alternative routes for the CA to reach the parenchyma. 18,34On the one hand, in case of large CA accumulation in the interstitial fluid, signal saturation and the slow transcytolemmal water exchange regime might have led to an underestimation of the tissue concentration, as during confinement of the CA in the blood pool.On the other hand, when interstitial CA accumulation is very weak and interstitial T 1 is above 850 msec, the interstitial signal remains nulled, masking the weak accumulation.For an initial tissue relaxation rate of 0.6 second À1 and a relaxivity of 3.6 mM À1 s À1 , T 1 < 850 msec is reached at a concentration above 0.16 mM, which constitutes the detectability limit.Below this CA concentration or for tissue T 1 > 850 msec, dynamic MRI with the chosen acquisition parameters is not suitable for quantification.If required for certain applications, the detectability limit can be lowered for instance by increasing the inversion time, taking care not to compromise the suppression of the tissue and blood signal.
Quantitative comparisons between in-vivo and ex-vivo modalities are affected by methodological difficulties.Parenchymal CA accumulation is expected to have continued after the last MRI acquisition and at blood flow arrest due to tissue hypoxia. 15Brain sampling and subsequent tissue processing are prone to partial tissue loss; eg, loss of small parts of the highly CA accumulating olfactory bulb can affect the average gadolinium quantification.Manual segmentation is subjective particularly in areas affected by partial volume effects, and exact correspondence of segmented structures with actually sampled tissue is uncertain, eg, for nerve roots, CSF in lateral recesses of the 4th ventricle, and the olfactory nerve layer.
Last but not least, the tissue CA relaxivity might be >3.6 mM À1 s À1 , i.e., beyond the range of estimates inferred from in-vitro studies, 27,28 as suggested by the comparison of MRI and ICP-MS gadolinium quantification for unperfused mice.The validation of the quantitative aspect of the trMRI approach over time would be a typical application of simultaneous PET-MRI acquisitions in conjunction with a bimodal imaging agent.
Bland-Altman analysis revealed reasonable agreement between gadolinium quantification by MRI and ICP-MS and confirmed a systematically higher value with MRI for perfused mice.Even for unperfused mice, the bias is still 0.11 mM (approximately 20% of the average gadolinium concentration).However, this analysis is limited by the small number of paired measures (a single measure per mouse, N = 21).Regional Gd quantification by ICP-MS has not been performed.This exploratory study revealed that the cerebral Gd-DOTA concentration is beyond the sensitivity limits of ICP-MS even in control mice, thus future studies will include quantification of the gadolinium content in subregions of the brain for correlation with MRI estimations.Additional studies are needed to distinguish whether CA accumulation results from extravasation or whether transport via the perivascular pathways of the glymphatic system is predominant.

Conclusion
We demonstrated that quantitative mapping of CA uptake in brain upon ip administration is feasible by time resolved MRI without pharmacokinetic modeling.The ip administered CA reaches the cerebral vasculature within 5 minutes and can be a minimally invasive alternative to iv administration.The ip administration route considerably lengthens the observation time.
= contrast agent; MRA = MR angiography; PD = proton density; trMRI = time resolved MRI; RARE = rapid acquisition with relaxation enhancement spin echo; VTR = variable repetition time; FLASH = fast low angle shot gradient echo; FOV = field of view; TE = echo time; TI = inversion time; TR echo = inter echo repetition time; NA = number of averages; NR = number of repetitions; T 10 = longitudinal relaxation time constant prior to CA administration.For all sequences, TE was minimized.

FIGURE 1 :
FIGURE 1: Timeline of the MRI acquisition protocol taking approximately 2 hours 40 minutes per animal.The time point of Gd-DOTA administration during the time resolved MRI acquisition was set to zero.MRA = MR angiography; PD = proton density; trMRI = time resolved MRI.

FIGURE 2 :
FIGURE 2: Representative axial T 1 map in milliseconds of mouse brain tissue in-vivo at 7-T.The white pixels represent failed fit.

FIGURE 3 :
FIGURE 3: Comparison between in-vivo Gd-DOTA estimation by MRI and ex-vivo gadolinium quantification by ICP-MS.For correlation analysis, the MRI measures were made in brain tissue including (a) and excluding (b) the ventricular CSF.The correlation coefficients remained unchanged when excluding the ventricles.The result of nonparametric correlation is shown because the two groups of perfused mice (N = 7) and unperfused mice (N = 5) treated with mannitol did not pass the normality test.When grouping all data for perfused mice regardless of treatment (N = 16), the correlation was significant (Spearman r = 0.8206, Pearson r = 0.8407, P < 0.001 for both).The gadolinium concentration by ICP-MS obtained on unperfused brains was close to the line of identity.The error bars show the standard deviation (SD) of the ICP-MS measures for four gadolinium isotopes.The MRI measures have no error bars because averaging over all brain regions produced standard deviations too high and standard errors too small to be displayed.(c) Bland-Altman plot showing bias and 95% limits of agreement separately for perfused mice (0.35 mM, 0.24 to 0.47 mM, N = 16) and for unperfused mice (0.11 mM, 0.03 to 0.20 mM, N = 5).An r 1 of 3.6 mM À1 s À1 was used for conversion of the trMRI signal.

FIGURE 4 :
FIGURE 4: Spatiotemporal distribution of Gd-DOTA accumulation in brain tissue and ventricular CSF.Maximum intensity projections (lateral and ventral views) generated from the 4D concentration maps show the spatial distribution of the contrast agent at three time points of the observation interval for two control mice and one mannitol treated mouse showing different degrees of uptake.OB = olfactory bulb; CW = circle of Willis; CB = cerebellum; GCVG = great cerebral vein of Galen; RRV = rostral rhinal vein; LV = lateral ventricles; 3V = 3rd ventricle; 4V = 4th ventricle.The nine corresponding animated gif files are available as supplementary figures (Figure S3).

TABLE 1 .
MRI Acquisition Parameters

TABLE 2 .
Analysis of MRI Acquisitions