Diffusion microscopic MRI of the mouse embryo: Protocol and practical implementation in the splotch mouse model

Purpose Advanced methodologies for visualizing novel tissue contrast are essential for phenotyping the ever‐increasing number of mutant mouse embryos being generated. Although diffusion microscopic MRI (μMRI) has been used to phenotype embryos, widespread routine use is limited by extended scanning times, and there is no established experimental procedure ensuring optimal data acquisition. Methods We developed two protocols for designing experimental procedures for diffusion μMRI of mouse embryos, which take into account the effect of embryo preparation and pulse sequence parameters on resulting data. We applied our protocols to an investigation of the splotch mouse model as an example implementation. Results The protocols provide DTI data in 24 min per direction at 75 μm isotropic using a three‐dimensional fast spin‐echo sequence, enabling preliminary imaging in 3 h (6 directions plus one unweighted measurement), or detailed imaging in 9 h (42 directions plus six unweighted measurements). Application to the splotch model enabled assessment of spinal cord pathology. Conclusion We present guidelines for designing diffusion μMRI experiments, which may be adapted for different studies and research facilities. As they are suitable for routine use and may be readily implemented, we hope they will be adopted by the phenotyping community. Magn Reson Med 73:731–739, 2015. © 2014 The Authors. Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the originalwork is properly cited.


INTRODUCTION
Worldwide collaborative efforts (1,2) are underway to develop a comprehensive functional annotation of the mouse genome, which will serve as a rich resource for elucidating human gene function. At least one third of all proteinencoding genes in the mouse genome are essential for development (3), and mutagenesis programs have identified many genetically engineered mutations that result in embryonic lethality (4). As complete inactivation of these genes will preclude further investigation in adult mice, a dedicated mouse embryo screening pipeline is urgently needed to characterize the anatomical phenotype of the increasing number of mutant embryos being generated (5). Furthermore, a dedicated screening pipeline will facilitate general mouse embryo phenotyping studies, such as investigations of the genetic causes of congenital abnormalities, and provide primary evidence to justify detailed follow-up studies. This will give insight into the functional and structural consequences of gene inactivation in developmental processes, as well as better models of human diseases and novel therapies.
Imaging platforms that enable detailed phenotyping include optical projection tomography (OPT) (6,7), microscopic computed tomography (mCT) (8,9), microscopic MRI (mMRI) (10,11) and high-resolution episcopic microscopy (HREM) (12,13). These modalities will most likely form the backbone of the screening pipeline as a complement to conventional optical histological examination, enabling acquisition of digital, high-resolution, three-dimensional (3D) data. To supplement this pool of techniques, advanced methodologies that enable visualization of novel tissue contrast will be essential for in-depth screening and assessment of specific pathologies.
In particular, mMRI has been widely used in embryo phenotyping studies since its first application in 1986 (14), offering high-throughput and nondestructive screening capabilities (15)(16)(17), 3D isotropic datasets with high resolution [18 mm (18,19)], and morphometric methods of analyses (20)(21)(22). While conventional relaxation-based MRI approaches may be used to identify structural malformations (16,20,23), they provide limited tissue contrast in the developing mouse brain, largely due to the lack of myelination (24)(25)(26). MRI contrast agents are commonly incorporated in the tissue fixation process to enhance contrast, boost the signal-to-noise ratio (SNR), and reduce scanning times by significantly lowering the tissue T 1 relaxation time. For example, Gd-DTPA (gadolinium diethylene triamine pentaacetic acid) has been used in mouse embryo phenotyping studies to visualize vasculature (10,27) and gross anatomy (16,19,20,23,28), and has been shown to provide some discrimination of developing white-matter structures (20). However, the specificity and sensitivity of the wide range of available agents have yet to be fully explored in the embryo (29).
Diffusion MRI is a powerful methodology (30) for noninvasively probing tissue microstructure. As the diffusion MRI signal is dependent on the dispersion pattern of water molecules, it is sensitive to the presence and orientation of ordered tissue structures. Thus, it can provide unique neuroanatomical information about the premyelinated embryo central nervous system (CNS), such as discrete delineation of multiple cortical layers, gray matter structures, and various white matter tracts (24)(25)(26). This has enabled volumetric analysis of whole brain and substructures (31), and facilitated 3D mapping of gene expression data (25). Accordingly, diffusion MRI, and in particular diffusion tensor imaging (DTI), has contributed to the characterization of normal brain development at multiple stages (21,26,31,32) and enhanced phenotypic assessment of mutant embryos (33)(34)(35).
Although DTI has been used successfully to assess the phenotypic characteristics of mutant embryos in several studies, practical implementation of this technique in an embryonic screening pipeline and widespread routine use of DTI for embryo phenotyping have been restricted by an extended scanning time-an order of magnitude higher than conventional structural imaging-which is partly due to limited methodological investigation. DTI studies in the developing mouse brain have typically minimized the scanning time by using 16 or fewer unique sampling directions. However, a Monte Carlo simulation has shown that the ideal number of directions is at least 30 (36), and under-sampling could compromise the robustness of common DTI measures, including the principal eigenvector, mean diffusivity (MD) and fractional anisotropy (FA). This may also preclude the use of advanced diffusion MRI techniques, such as spherical deconvolution (37)(38)(39), super resolution (40,41), and the apparent fiber density (42). Furthermore, there is currently no established protocol for mouse embryo phenotyping using mDTI that ensures optimal data acquisition for any user. An approach that can accommodate different needs and MRI hardware systems would be useful for providing more robust comparisons of data between research facilities, such as the MD and FA, which are dependent on the scanning parameters (e.g., diffusion time). For example, there is a large variation in the experimental set-up to control sample temperature, whether passive, intrinsic, or extrinsic, as the gradient subsystem set-up varies in the diameter of the unit and efficiency of the gradient cooling. Thus, these factors need to be considered during the experimental design stage.
We present two protocols for phenotypic characterization of embryos using diffusion mMRI, which provides users with an experimental procedure for their particular diffusion mMRI study requirements, MRI system hardware and experimental design. Building on previous work (23)(24)(25)(30)(31)(32)(33)(34), we have developed protocols that enable data acquisition in a reasonable timeframe with a standard imaging sequence, such that this technique may be readily implemented in an embryonic imaging pipeline and by individual researchers. As the protocols initially account for the choice of diffusion protocol and MRI hardware, which can affect the embryo preparation by means of sample heating and confound resulting data, it provides adaptability and flexibility for different phenotyping studies and research facilities.
As a demonstration, we also implement our protocols in an example application: a whole-body phenotyping study that examines the wider effects of spina bifida on the developing CNS. In particular, we investigate the splotch mouse model of human neural tube defects (NTDs) (43). NTDs are a common group of congenital malformations, including spina bifida and exencephaly, which affect 0.5-2 per 1000 pregnancies worldwide (44). These mice carry a mutation in the Pax3 (paired box 3) gene, a transcription factor implicated in these conditions (45,46). For this example application, we applied a two-stage experiment: a short diffusion protocol and a more extensive detailed diffusion protocol. The short diffusion protocol uses six directions at 75 mm isotopic, providing a 3 h data acquisition time per embryo, such that it may be used for preliminary investigations. More accurate DTI measures [35] were acquired for detailed imaging using an extensive detailed diffusion protocol, which used 42 directions at 75 mm isotopic, providing a 19 h data acquisition time per embryo.

Diffusion mMRI Investigations: Short and Long Diffusion Protocols
Based on our experience, we have developed methods to enable a two-stage experimental procedure for implementing diffusion mMRI of mouse embryos (Fig. 1), which provides preliminary and/or detailed diffusion data. The first step is to select a short diffusion protocol (short diffusion weighted [DW] protocol) for relatively fast preliminary imaging (e.g., six directions at a defined b-value). Subsequently, the MR imaging parameters (e.g., repetition time [TR]) and experimental setup are then empirically optimized to ensure that the protocol does not cause sample heating and maximizes SNR, while ensuring image quality. Contrast agent concentration and fixative washout duration is then selected to provide the optimal combination of T 1 and T 2 for the desired protocol, experimental setup and MRI hardware. Upon implementation of the short DW protocol and assessment of the data, a more extensive diffusion protocol (long DW protocol) may be adopted for more detailed investigation of embryos. The imaging parameter optimization process is repeated before implementation if a different diffusion weighting is used.

Embryo Preparation
All animal work was performed according to protocols approved by the UK Home Office. Time-mated wild-type (C57BL/6) mice were killed at 15.5 days post coitum (dpc) by cervical dislocation, and the embryos were dissected from the uterus. The umbilical cords were cut close to the placenta allowing the embryos to exsanguinate in Hanks' solution at 37 C. A small incision ($ 2 mm) was made in the side of each embryo, just below the liver, to facilitate penetration of the contrast agent. The embryos were immersion-fixed in a solution of 4% PFA (4% aqueous formaldehyde from paraformaldehyde) and stored at 4 C.

Data Acquisition
A 9.4 Tesla (T) magnetic resonance imaging system (Agilent Technologies Inc., Palo Alto, CA) was used with a 1T/m gradient set (60 mm diameter, with a water-based gradient cooling unit, set to 17 C) and 26 mm volume coil (RAPID Biomedical GmbH, W€ urzburg, Germany). For scanning purposes, each embryo was placed in a sealed 10-mL syringe filled with Perfluorosolv PFS-1 solution (Solvay Solexis Inc., West Deptford, NJ), which prevents dehydration and provides susceptibility matching. Each embryo was carefully secured in place with the plunger, using gauze for additional padding. The sample temperature was monitored throughout all acquisitions using an MR-compatible temperature probe designed for physiological monitoring (SA Instruments Inc., Stony Brook, NY).

Tissue Temperature Equilibrium
Preliminary experiments were conducted with and without the MRI diffusion weighting gradients to determine the main source of sample heating. As diffusivity changes with temperature (47), the temperature of an unstained embryo was monitored at different TR values during diffusion-weighted (DW) imaging to determine the lowest value of TR that produced a sample temperature of 19 C with a change of no more than 1 C. To verify this, we chose to investigate the heating effects by applying the diffusion weighting gradients along the read, phase and slice directions individually rather than simultaneously. The diffusion weighting was identical to the subsequent imaging experiment (

Contrast Agent Concentration Optimization
To minimize steady state effects and ensure that the magnetization was near full relaxation, a tissue T 1 value of approximately one fifth of the TR value was desired. To determine the contrast agent concentration that would provide this value, the T 1 values from a previously acquired dataset (23) of CD-1 embryos stained with 4-16 mM Gd-DTPA were initially used as a benchmark. Building on this experience, splotch wild-type embryos (Pax31/1) were immersion-fixed in 4% PFA doped with 2 mM Gd-DTPA (Magnevist, Bayer-Schering, Newbury, UK) (n ¼ 1 per concentration), and stored at 4 C for 2 weeks. The T 1 values of these embryos were measured approximately during preliminary scans using inversion recovery 3D spin-echo acquisitions.

Fixative Washout
Fixative washout was investigated as a possible method for increasing the SNR, as this has been shown to increase the tissue T 2 with a much smaller increase on the tissue T 1 (48). The embryo stained with 2 mM Gd-DTPA was immersed in phosphate buffered saline (PBS)

MRI Sequence (3D FSE versus 3D SE)
Two widely available DW sequences, 3D FSE and 3D SE, were compared to assess which sequence provides the most appropriate combination of SNR, scan time and image quality. The same embryo (stained with 2 mM Gd-DTPA) was used at five and seven weeks after immersion in PBS for the 3D SE and 3D FSE scans, respectively. The resolution, TR and diffusion parameters were set to be the same (75 mm isotropic, matrix ¼ 200 Â 96 Â 120, FOV ¼ 15 Â 7.2 Â 9 mm, TR ¼ 500 ms, NSA ¼ 1, G ¼ 0.5 T/m, d

Phenotypic Assessment
Using our optimized experimental procedure, we acquired DW images of a splotch homozygous embryo (Pax3 Sp2H/Sp2H ; on C57BL/6 background) and a wild-type littermate (both immersion-fixed in 4% paraformaldehyde doped with 2 mM Gd-DTPA, followed by 2-week immersion in PBS doped with Gd-DTPA). The 3D DW images were corrected for movement with a rigid registration with three degrees of freedom (translation only) using the FLIRT registration tool (49,50), which is part of the FMRIB Software Library (FSL v 4.1, Oxford, UK). Direction encoded color (DEC) maps were computed and visualized for comparison using MedINRIA (v1.9.4, Asclepios Project, France). For comparison with histology, the embryos were wax-embedded, sectioned into 10mm sagittal slices and stained with hematoxylin and eosin (H&E). Additionally, reference images from the Schambra mouse embryo brain atlas (51) were used.

Tissue Temperature Equilibrium
Our preliminary experiments indicated that the MRI diffusion weighting gradients provided the main source of sample heating (temperature rises were only apparent for diffusion-weighted scans). We determined that a minimum TR value of 500 ms was needed to maintain a stable sample temperature, which was important as diffusivity changes with temperature (47). Figure 2 shows that the temperature variation during test scans with identical imaging and diffusion weighting (G ¼ 0.5 T/m, d ¼ 3.5 ms, D ¼ 8 ms, b-value ¼ 1498 s/mm 2 ) (applied sequentially in the read, phase and slice directions) stayed within 19 6 1 C. Using this TR value for our short and long DW protocols (6 and 42 directions, respectively), the sample temperature remained stable throughout all imaging acquisitions (19 6 1 C).

Contrast Agent Concentration Optimization
Using preliminary measurements of stained splotch wild-type embryos and previously acquired T 1 data of stained CD-1 embryos (23), we determined that 2 mM was the optimal concentration of Gd-DTPA as it provided tissue T 1 values (109 ms in the cortex and 104 ms in the spinal cord) that were approximately one fifth of the TR value (500 ms), which was necessary for minimizing steady state effects and ensuring the magnetization was fully relaxed.

Fixative Washout
Following 2-week immersion in PBS doped with Gd-DTPA, we observed an approximate three-fold increase in SNR. This was attained for each of the unweighted measurements (average SNR from 63 to 163) and each of the 42 directions (average SNR from 40 to 98). Additionally, PBS immersion resulted in an approximate two-fold increase in the T 2 values of the spinal cord and cortex, while the T 1 values remained relatively constant (within 95% confidence intervals) ( Table 1).

MRI Sequence (3D FSE versus 3D SE)
We observed a higher SNR per unit time (approximately three-fold) using the 3D FSE sequence compared with the 3D SE sequence for both the unweighted measurements (average 58 versus 16) and 42 directions (average 31 versus 10). A comparison of line profiles between 3D SE and 3D FSE data indicated that the level of blurring in 3D FSE scans was acceptable (Supporting Fig. S1, which is available online). On visual inspection, there was very limited blurring in the DEC map generated from 3D FSE data compared with 3D SE data (Figs. 8c and 3, respectively).

Phenotypic Assessment
Using our short DW protocol (6 directions), we detected defects in the spinal cord of the splotch mutant embryo (Fig. 4) and a reduction in the FA of its limb muscles ( Table 2) when compared with the wild-type embryo. Additional differences in other areas of the central nervous system were indicative, prompting further investigation using a long DW protocol (42 directions).
Our long DW protocol (42 directions) enabled clear visualization of several anatomical brain regions according to diffusion directionality in the wild-type embryo, including the spinal cord, cortex, and thalamus (Fig. 5). The DEC maps of the wild-type embryo indicated that the fibers of the marginal zone (MZ) and mantle layer (ML) of the spinal cord were oriented in the ventral-dorsal and rostral-caudal directions, respectively, which correlated with the histology (Fig. 6). Furthermore, the DEC maps elucidated rich microstructural information that is not evident using conventional T 2 *-weighted MR imaging. For example, the thalamus appears homogenous in a representative T 2 *-weighted MR image (23) (Fig. 7c), whereas the DEC maps revealed contrast within this structure (Fig. 7a) that corresponded to the lateral and posterior thalamic nuclei and the reticular thalamic nucleus (white and yellow arrows, respectively) in a histological slice [Fig. 7b, modified from the Schambra atlas (51)].
Application of our methodology to the splotch mutant enabled phenotypic assessment of the whole embryo (Fig. 8). The primary spinal cord defect (spina bifida) could be clearly visualized in the DEC maps, and differences were evident in the overall brain size. Regions demarcated by the principal eigenvector, such as the pons, hindbrain, and midbrain, were readily identifiable, while the regional delineation in the thalamus was not as apparent in the mutant as the wild-type. Strikingly, a clear herniation of the brain stem into the vertebral canal was visible, indicating the presence of the Chiari II malformation, which is a common finding in humans with spina bifida, but rarely described in previous studies of genetic mouse NTD models. Additionally, abnormal limb muscles were evident in the mutant compared with the wild-type, as demonstrated by the reduction in FA ( Table 2).

DISCUSSION
In this study, we present two protocols for mouse embryo diffusion mMRI, together with guidelines for embryo preparation and imaging setup, which can be readily implemented to achieve a reasonable timeframe for data acquisition, making it suitable as a supplementary methodology for mouse embryo imaging and hopefully enabling more widespread use within the phenotyping community. We have demonstrated that DTI can provide sensitivity to tissue microstructure in the developing wild-type mouse brain that is not  apparent using conventional T 2 * relaxation-based MRI approaches, in agreement with previous studies (24)(25)(26)(27)30,32,34,35). Subsequently, we successfully applied our protocols for investigation of the splotch mouse model of human NTDs, which enabled the first DTI assessment of spinal cord pathology in this model. Our findings suggest that microstructural differences between phenotypes may be elucidated using appropriate diffusion MRI protocols, particularly in the CNS and skeletal muscles.
To date, application of DTI for mouse embryo phenotyping has been limited due to the extensive data acquisition time involved in this methodology. While some embryo phenotyping studies have been carried out using DTI (33)(34)(35), widespread use will require guidelines for an approach that enables higher-throughput screening and adaptability for different studies and research facilities. Our development brings us closer to this aim, allowing the user to balance the scan time, resolution, SNR, and b-value according to their needs. For example, in our example study we chose a relatively low b-value (1498 s/mm 2 ) to reduce the total imaging time and maintain high resolution and SNR. Similarly, another study has recently been designed and optimized for ex vivo DW imaging of pig brains (52).
A wide array of diffusion mMRI protocols have been used in previous embryo phenotyping and developmental studies (24,(31)(32)(33)(34)(35), and there is currently no established methodology that ensures optimal data acquisition for any user. DTI data have typically been acquired using a 3D DW multiple SE sequence (53) in $ 15-30 h with 7-14 directions and 2-4 signal averages at $ 62 Â 83 Â 83 mm to 133 Â 131 Â 100 mm (12.5 dpc to adult mice). A 3D SE sequence has also been used providing DTI data in $ 32 h with seven directions and four signal averages at $ 78 Â 94 Â 94 mm (15.5 dpc) (26). More recently, Aggarwal et al implemented a 3D DWgradient and spin echo (GRASE) sequence (25) that enabled acquisition of DTI data in 1 h 50 min per direction at 60 mm isotropic (12 dpc).
Our example implementation produced an experimental procedure that enabled acquisition of DTI data in 24 min per direction at 75 mm isotropic using a standard 3D FSE sequence, such that preliminary or detailed screening of mid/late-gestation embryos was possible in $3 h (with 6 directions plus one unweighted measurement) and $19 h (with 42 directions plus six unweighted measurements) per embryo, respectively, with no detectable instabilities introduced. While 3D SE sequences have been used more commonly in previous DTI studies, we found that a 3D FSE sequence provided higher SNR per unit time. A disadvantage of using FSE compared with SE is that blurring may be introduced. However, our protocols led to acceptable levels of blurring as the signal amplitudes from the first and last echo in the FSE echo train were comparable and we were able to shim the samples well. For this study, we also compared 3D-SE and 3D-FSE imaging modules as these are commonly available on preclinical systems. We used 3D FSE in our mDTI experimental procedure because of the comparable image quality and higher SNR per unit time compared with 3D SE. However, higher SNR per unit time sequences, such as GRASE (25) and SSFP (54,55), could potentially be used with prior understanding of their possible artifacts.
Before data acquisition, we recommend optimization of several parameters, including the TR value (for tissue temperature equilibrium), contrast agent concentration (for optimal tissue T 1 values) and the time required for fixative washout (for improved SNR). Earlier embryo phenotyping and developmental studies using DTI (24,26,32,34,35) immersion-fixed embryos in a tissue fixative without a contrast agent, some of which (24,32,35) used fixative washout by means of PBS immersion for at least 24 h before imaging. More recent embryo studies have incorporated 0.1 mM (31) and 1 mM (25) Gd-DTPA in the tissue fixation process (for more than 48 and 24 h, respectively), with no fixative washout. In our example, we determined that 2 mM of Gd-DTPA followed by PBS immersion for 2 weeks was optimal for the experimental procedure required in our specific study.
The DEC maps acquired using our protocols provided microstructural information that was not apparent using conventional relaxation-based MRI (Fig. 6), demonstrating the value of mDTI as a supplementary methodology for mouse embryo imaging. Visualization of the spinal cord tissue structure and thalamus using the DEC maps corresponded well with histology (Figs. 5 and 6), indicating that our methodology provided anatomically relevant information.
We were also able to detect additional differences within the mutant brain (Fig. 8), opening up new avenues for further investigation. For example, while studies of mouse genetic models of NTDs have been able to replicate many of the clinically observed defects, including spina bifida, craniorachischisis, and exencephaly (44), there have been very few reports in these mice of the Chiari type II malformation, in which the brainstem is elongated and the cerebellar vermis herniates through the foramen magnum. This disorder is commonly associated with spina bifida in humans (56), and yet the pathogenic basis of this link remains poorly understood. Our finding of the Chiari II malformation co-existing with the open spina bifida in the genetic splotch mouse model offers a new opportunity to investigate the developmental basis of this malformation association. In the regions outside the CNS and skeletal muscles, we observed subtle DEC color differences, which is possibly due to gross differences in the orientation and position of regional anatomy of the organ systems.
With the creation of thousands of mutant mouse embryos facing researchers in the coming years, the challenge of phenotyping will require a sophisticated approach that uses existing and emerging techniques available in a complimentary way to maximize the information obtained during this "golden" period (57). Amongst these techniques, diffusion mMRI has the potential to offer unique information about tissue microstructure and connectivity in three dimensions. For example, a clinical study has shown that frontal white matter diffusion abnormalities may be detected using DTI, enabling diagnosis of mild traumatic brain injury in patients, which is not possible using MRI or CT (58). Thus, diffusion mMRI has the potential to reveal novel phenotypic characteristics related to microstructural changes that are not available using other techniques. However, it will be essential to compare the DTI data with existing and emerging techniques (57) to assess sensitivity and specificity to pathology and gauge its contribution to the field.
As demonstrated by our initial example, the protocols would enable users to develop an experimental procedure specific to their study and MRI system hardware. For example, shorter length scales could be probed using optimized gradient waveform SE sequences (59,60), or more efficient gradient and sample cooling systems, which would allow a reduction in TR or higher b-value. Also, studies at earlier developmental stages could allow for simultaneous imaging of multiple embryos owing to their smaller size, thus increasing the throughput. Further improvements could be made through the use of a dedicated mouse embryo imaging volume coil, which would provide better SNR.

CONCLUSIONS
We have developed two protocols for whole-body diffusion mMRI, which provided DTI data in 24 min per direction at 75 mm isotropic resolution, enabling preliminary and detailed screening in $ 3 h (6 directions plus one unweighted measurement) and $ 19 h (42 directions plus six unweighted measurements) per embryo, respectively. We determined that for our study, fixation with Gd-DTPA followed by tissue rehydration enabled a combination of T 1 reduction without excessive T 2 shortening, providing increased SNR. Application of our protocols to the splotch mouse model of human NTDs provided assessment of the microstructural differences of the whole CNS between the wild-type and mutant mouse. As our protocols can be readily applied to embryo imaging studies using a range of MRI hardware configurations, it may enable implementation of diffusion mMRI as a supplementary technique in mouse embryo screening pipelines and facilitate greater use within the phenotyping community. With the development of methods for automated analysis of DTI-derived data (32), diffusion mMRI could prove a powerful tool for routine embryo phenotyping. Thus, we hope our guidelines will enable a wider adoption of diffusion mMRI for embryo phenotyping studies, which may provide new insights to congenital disease models.