µMRI–HREM pipeline for high-throughput, high-resolution phenotyping of murine embryos

Abstract

Rapid and precise phenotyping analysis of large numbers of wild-type and mutant mouse embryos is essential for characterizing the genetic and epigenetic factors regulating embryogenesis. We present a novel methodology that permits precise high-throughput screening of the phenotype of embryos with both targeted and randomly generated mutations. To demonstrate the potential of this methodology we show embryo phenotyping results produced in a large-scale ENU-mutagenesis study. In essence this represents an analysis pipeline, which starts with simultaneous micro-magentic resonance imaging (µMRI) screening (voxel size: 25.4 × 25.4 × 24.4 µm) of 32 embryos in one run. Embryos with an indistinct phenotype are then cut into parts and suspect organs and structures are analysed with HREM (high-resolution episcopic microscopy). HREM is an imaging technique that employs ‘positive’ eosin staining and episcopic imaging for generating three-dimensional (3D) high-resolution (voxel size: 1.07 × 1.07 × 2 µm) digital data of near histological contrast and quality. The results show that our method guarantees the rapid availability of comprehensive phenotype information for high numbers of embryos in, if necessary, histological quality and detail. The combination of high-throughput µMRI with HREM provides an alternative screening pipeline with advantages over existing 3D phenotype screening methods as well as traditional histology. Thus, the µMRI-HREM phenotype analysis pipeline recommends itself as a routine tool for analysing the phenotype of transgenic and mutant embryos.

Introduction

Our understanding of the aetiology of hereditary diseases depends on characterizing the function of genes and their products in embryogenesis. Knowledge of the underlying genetics of inherited disorders will form a basis for developing early diagnostic tools and novel therapies.

As fundamental developmental mechanisms are conserved across mammals, genetically modified mouse lines carrying targeted mutations are widely used to investigate individual gene function. Mutant mouse lines also provide experimental models for understanding profound genetic rearrangements, such as chromosomal deletions, rearrangements or translocations that are responsible for many inherited disorders. Large-scale mutagenesis screens are now being performed in order to identify potential mouse models for the many inherited human disorders whose genetic origins are yet unknown (Nolan et al. 2000; Yu et al. 2004; Cordes, 2005; Shen et al. 2005).

The essential first step in analysing individual mutant or transgenic organisms is a comprehensive description of the phenotype. Intrauterine death of the homozygote offspring of mutant lines often leads to the necessity to analyse the phenotype of embryos as soon as the development of all vital organ systems is complete. Methods permitting high-throughput and high-resolution morphological analysis of embryos are therefore urgently needed. However, none of the currently available imaging methods alone is able to meet all the requirements for detailed large-scale phenotypic screens of embryos. Despite technological advances, computed tomography (CT), magnetic resonance imaging (MRI) and ultrasound do not provide high enough resolution or tissue contrast for analysing organ details (Shen et al. 2005; Heinzer et al. 2006; Johnson et al. 2006). Techniques using tissue fluorescence, such as confocal microscopy, block surface imaging, episcopic fluorescence image capturing (EFIC) and optical projection tomography (OPT) (Ewald et al. 2002; Sharpe et al. 2002; Weninger & Mohun, 2002) do not permit precise visualization of tissue architecture.

Histological section-based methods can provide three-dimensional (3D) information on tissue architecture, but these techniques are labour intensive and time consuming and the quality of the 3D models is low due to inexact section alignment or the use of relatively thick sections. (Weninger et al. 1996; Streicher et al. 1997, 2000; Soufan et al. 2003).

As a compromise, micromagnetic resonance imaging (µMRI) was recently established as a routine method for high-throughput phenotypic analysis of embryos (Bamforth et al. 2004; Schneider & Bhattacharya, 2004; Schneider et al. 2004a, b; Bogani et al. 2005). It facilitates fast morphological screening of organs and organ systems for multiple embryos in a single run. However, the resolution (voxel size 25.4 × 25.4 × 24.4 µm) is too low to distinguish different cell types and tissues. Therefore, µMRI screening might miss small but essential morphological abnormalities. Diagnosis of small lesions and tissue abnormalities requires methods that provide data of histological quality. High-resolution episcopic microscopy (HREM), a recently developed episcopic imaging technique, is such a method (Weninger et al. 2006). In contrast to early episcopic methods (Hegre & Breshear, 1946, 1947; Odgaard et al. 1990; Odgaard, 1993; Weninger et al. 1998; Weninger & Mohun, 2002), HREM rapidly provides volume data sets in almost histological quality in voxel sizes down to 0.4 × 0.4 × 1 µm. However, HREM cannot process multiple specimens simultaneously in high-throughput fashion.

Here we present a new screening method that permits high-throughput, high-detail phenotype analysis of transgenic and mutant mouse embryos. It essentially represents a systematic screening pipeline combining modified µMRI and HREM imaging techniques. Embryos are first screened by µMRI. Embryos with a pathological phenotype detected by µMRI are subsequently analysed by HREM. This analysis pipeline quickly results in highly precise 3D information of tissue architecture and small detail in the context of gross morphology and overall embryo anatomy.

Materials and methods

Mouse embryos arising from a genome-wide phenotype-driven recessive screen to identify developmental malformations were studied at 15.5 days post-conception (dpc) as most major organs can be clearly identified at this stage (Kaufman, 1995). Older embryos can be analysed without major changes in resolution and contrast (Schneider et al. 2004a, b). Embryos were immersion-fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) containing 2 mm of the unspecific contrast agent dimeglumine gadopentetate (Gd-DTPA, Magnevist) for several days at 4 °C.

Gd-DTPA is a clinically widely used paramagnetic, extracellular fluid agent with a relatively short residence time in the vascular system. It does not stain tissues specifically or directly bind to proteins, carbohydrates or lipids, but changes the longitudinal relaxation time T₁ locally. The efficacy of this relaxation mechanism increases with the mobility of the hydrogen protons (Caravan et al. 1999). Given that the applied sequence provides strong T₁-contrast, the signal read out by the sequence is, for example, stronger in ventricular cavities than in the myocardium.

The penetration with Gd-DTPA by means of diffusion depends principally on skin properties, embryo size and fixation time (Schneider et al. 2004a, b). In our experience a fixation time of 4 days represents a good compromise between image quality, contrast reproducibility and practicality.

The position of the embryo within the radiofrequency (RF)-probe also influences contrast. Although the RF-coil provides a homogeneous RF-field over 60–70% of the sample, the top and bottom layers are affected by reduced homogeneity, which has an impact on contrast. A detailed description of the µMRI technology used here can be found in previous publications, to which the reader is referred (Schneider et al. 2003a, b, 2004a, b).

The fixed embryos were embedded in 1% agarose in water with 2 mm Gd-DTPA in 28-mm multi-embryo imaging glass tubes (32 embryos per tube). The tubes underwent µMRI scanning at a resolution of 25.4 × 25.4 × 24.4 µm per voxel in overnight runs, imaging all 32 embryos simultaneously. µMRI images were archived as TIFF files. The volume data sets were visualized with the software package Amira 4.0 (Mercury Systems) and analysed using virtual re-sectioning. In selected embryos the heart and larger blood vessels were manually reconstructed using the label field function of the software. Surface rendered 3D models were generated for presentation.

Following µMRI, embryos were stored in neutral buffered formaldehyde. In this medium, they were segmented into three parts (head, neck/thorax and abdomen/pelvis) and prepared for HREM data generation (Weninger et al. 2006). Each embryo part was washed in tap water (2 h) and then dehydrated in ethanol (30, 50, 70, 80, 90 and 96%, each for between 30 min and 1 h) in order to extract the majority of the Gd-DTPA from embryo tissue. The embryo parts were then infiltrated with, and subsequently embedded in, resin (Immuno-Bed, Polyscience Inc.) dyed with ‘eosin spritlöslich’ (0.4 mg 100 mL⁻¹, Waldeck GmbH). The resin blocks, including the specimens, were mounted and sectioned on a rotary microtome (CUT 4060E, Microtec). During fully automated sectioning, episcopic images of each freshly cut block surface were captured using a digital video camera (Leica DFC 480) sitting on the phototube of a compound microscope (Leica ‘DM LM’). The optical pathway contained a YFP filter cube (excitation filter 500/20 nm, emission filter 535/30 nm) and was aligned with the upper stopping point of the block holder excursion way. Stacks consisting of 1200–1500 aligned digital images of subsequent block surfaces every 2 µm were obtained over a few hours of sectioning. The images are of comparable quality with those of digital images of true histological sections, but do not show inhomogeneous staining between single sections, sectioning or mounting artefacts. Tissue staining appeared homogeneous within each specimen (Fig. 1). Although the histological sections produced can be mounted on glass slides, they were removed because they were of no further use in this study. The specimens were analysed by scrolling through Quick Time movies (Quick Time 7.1.3) produced from the image stacks or converted to volume data (voxel size: 1.07 × 1.07 × 2 µm and 2.14 × 2.14 × 2 µm, respectively) and analysed with the software package Amira 4.0 (Mercury Systems). Although the orthogonal slice function of Amira was sufficient for analysis, we manually outlined the surface of the thyroid gland and the oesophagus in one selected embryo using the label field function of the software and produced surface-rendered 3D models for data presentation. No special hardware was required for QuickTime movie analysis. Analysing HREM data in Amira required the use of a high-end PC. This PC was equipped with 12 GB RAM and running Windows Professional 64-bit.

Fig. 1.

Virtual section planes through an HREM data volume. Note the uniformity of tissue staining in all parts of the embryo. Scale bar, 500 µm.

Results

We used our novel methodology (Fig. 2) to analyse the phenotype of mutant mouse embryos and wild-type littermates arising from an ENU mutagenesis study.

Fig. 2.

Flow chart of the phenotyping pipeline.

As expected, µMRI data quickly provided an overview of the morphology of all organ systems in all embryos scanned. Structural abnormalities, such as palatine clefts or ventricular septal defects, could be precisely diagnosed (Fig. 3). However, not all embryos showed malformations, which could be reliably characterized by µMRI screening. In a number of specimens µMRI data merely hinted at aberrant anatomical features, such as small ventricular septal defects, enlarged lymph sacs, thinning of the myocardium or right retro-oesophageal subclavian artery. To verify or discount these putative malformations, HREM analysis of the suspect embryo parts was performed. Because HREM analysis could focus on suspect embryo parts (e.g. chest, head, neck, abdomen), precise high-resolution 3D phenotype characterization was available within short data generation and data processing times (Fig. 4).

Fig. 3.

Examples of abnormal morphology detected by high-throughput µMRI and HREM of 15.5-dpc mouse embryos. (a–c) µMRI data showing a palatine cleft (Pc in a), nuchal oedema (arrows in b), and a ventricular septal defect (asterisk in c). (d–f) HREM data showing defects not detected by µMRI: large hemiazygos (Hv) and azygos (Av) veins (d), persistent vitelline vein segments proximal and distal to the liver (arrowheads) and an umbilical vein (Uv) running in the body wall (e), and a hypoplastic left lobe (lThy) of the thyroid gland (f). B, brain; Sc, spinal cord; Li, liver; C, heart; G, gut; Aa, aortic arch; Db, ductus arteriosus; Llv, lumen of left ventricle; Lrv, lumen of right ventricle; V, vertebra; A, aorta; K, kidney; L, lung; U, future bladder; D, diaphragm; P, pancreas; E, oesophagus. Scale bars, 500 µm.

Fig. 4.

Synergy of µMRI and HREM in high-throughput phenotyping of 15.5-dpc mouse mutants. (a–d) µMRI data (a,b) suggest a right retro-oesophageal subclavian artery – HREM data (c,d) confirm it due to higher spatial resolution (inlay, arrowhead). (e–h) µMRI data (e,f) suggest thinning of the right ventricle myocardium – HREM data (g,h) confirm it (inlay). (i–l) µMRI data show lumbar oedema (arrow in i) but no obvious cardiac defects (j) – HREM data (k,l) reveal a ventricular septal defect. Sc, spinal cord; Aa, aortic arch; E, oesophagus; T, trachea; Vs, ventricular septum; lv, left ventricle; rv, right ventricle. Scale bars, 500 µm.

Significantly, µMRI screens frequently revealed complex malformations involving multiple organ systems. On the basis of this information we were able to perform focused HREM analysis of organs barely accessible by µMRI due to its relatively low resolution, but usually co-affected in such malformations. This allowed, for example, the detection of thyroid gland defects in specimens in which µMRI had revealed exencephaly and heart defects (Fig. 3). In this way, the pipeline proved to be optimal for a fast and comprehensive description of complex phenotypes.

Often, the status of the embryo presenting itself on µMRI images indirectly suggested a cardiovascular defect by showing oedema or enlarged veins. However, µMRI resolution was too low to diagnose a cardiovascular malformation. In such cases selective high-resolution HREM analysis of the cardiovascular system was performed. In some embryos this revealed defects that were generally small but severe and of haemodynamic relevance. The spectrum we observed included absent ductus venosus, multiple small ventricular septal defects, persistent connection of the umbilical vein to the heart, persisting vitelline veins and a second inferior vena cava (Figs 3 and 4).

Discussion

Our data show the benefits of a systematic phenotype screening pipeline, combining µMRI and HREM. The speed of high-throughput µMRI imaging and its ability to screen multiple embryos simultaneously allows the rapid identification of structural abnormalities in a large number of specimens in a short time. The quality of the µMRI data is good enough to permit the identification of many structural defects and select embryos that warrant further high-resolution analysis. Because µMRI is a non-destructive imaging method and the tissue contrast enhancer Gd-DTPA can be washed from the tissues with water and low-concentration ethanols, µMRI-prescreened specimens can be re-analysed with high-resolution destructive sectioning techniques. HREM proved to be optimal for this. First, its spatial resolution is a minimum of 3–5 times higher than that obtainable using other high-resolution imaging methods such as OPT and external marker-based congruencing (EMAC) (Streicher et al. 1997, 2000; Sharpe et al. 2002). Secondly, HREM data are of almost true histological appearance, but in contrast to traditional histological section series, HREM information is produced more rapidly (no section preservation and section processing) and does not suffer from section, mounting or section processing artefacts. Thirdly, HREM, in contrast to fluorescence-based imaging techniques [confocal imaging, EFIC, block surface imaging and OPT (Sharpe et al. 2002; Ewald et al. 2002; Weninger & Mohun, 2002)] permits highly accurate analysis of tissue layers and organ boundaries (see http://www.meduniwien.ac.at/3D-Rekonstr/HREM/). Finally, HREM data generation is simple and cheap.

µMRI cannot diagnose small structural defects or tissue abnormalities. However, it is sufficient to identify embryos that are likely to show such defects. Along our pipeline suspect specimens are subjected to HREM analysis, which is able to confirm or discount the suspected morphological or histological abnormalities. Initial µMRI diagnosis allows us to focus HREM analysis on suspect organs and only relevant body parts are embedded for HREM processing. Although embryo segmentation is not absolutely necessary for HREM analysis of 15.5-dpc mouse embryos, we performed it for four reasons. First, commercially available embedding moulds are 6 mm deep and resin blocks larger than 8–10 mm have a tendency to break during the sectioning procedure. Secondly, small specimens can be dehydrated and embedded more rapidly than larger specimens. Thirdly, HREM data are very large, due to their high resolution, which complicates data handling. Therefore, using only the relevant parts of suspect embryos creates a comparably small data volume that can be processed and analysed more quickly. Fourthly, sectioning only embryo parts instead of entire embryos significantly reduces the data generation time. To avoid artefactual damage of the organs near the cutting planes, the exact location of each cutting plane is defined individually according to the location of the suspected malformation.

The costs of this screening pipeline are as follows. MRI scanning of 32 embryos is performed in overnight runs and costs 12 Euro per embryo. An experienced investigator can analyse the MRI data of a single embryo in 20 min. HREM data take 4–7 h per specimen and cost less than 1 Euro, although this sum does not include hardware costs (the price of the data generation apparatus used was approximately 50 000 Euros) or labour costs. Neither HREM itself nor the specimen analysis pipeline is commercialized. Therefore, a valid ‘all-inclusive’ price for HREM imaging has yet to be calculated.

Figure 2 shows the data analysis pipeline. It takes at least 7–8 days between Gd-DTPA staining and HREM data analysis, although several batches of 32 embryos can be processed simultaneously. In principle, a sequential pipeline can be started daily. In such a case, HREM data analysis has to be restricted to a maximum of two selected specimens per batch. Alternatively, several HREM data generation apparatuses can be operated simultaneously, although operating more than three HREM apparatuses is a major logistical problem mainly due to the enormous volume of data generated. Reducing the resolution of HREM data would significantly reduce the data volume as well as the time required to process a specimen. For example, section thicknesses of 4 µm instead of the routinely used 2 µm results in volume data with a voxel size of 1.07 × 1.07 × 4 µm and halves the raw data volume, and almost halves the data generation time. An additional reduction of the resolution of the 2D block surface image from 1.07 × 1.07 µm to 4.28 × 4.28 µm results in an almost cubic voxel size of 4.28 × 4.28 × 4 µm and in a raw data size, which is 32 times smaller than the data we routinely produce (voxel size 1.07 × 1.07 × 2 µm). But we consider this to be an unacceptable loss of resolution, as it precludes precise analysis of small blood vessels and tissue layers.

On average, the µMRI data of 32 simultaneously scanned embryos consists of 2000 TIFF files in 16-bit image depth. The size of each TIFF file is 2 Mb. Thus, one embryo is represented in approximately 125 Mb data. By contrast, HREM data usually consist of 500–2500 TIFF files in 8-bit image depth. The number of files depends on the size of the embryo parts analysed by HREM. Morphology and tissue architecture of relevant organs and organ systems can be sufficiently analysed in greyscale HREM images, and the size of one TIFF file produced in greyscale mode is 1.4 Mb. One 15.5-dpc mouse embryo, sectioned in 2-µm steps, is represented in approximately 7–9 Gb greyscale HREM data. Thus, HREM data are approximately 65 times larger than µMRI data. We suggest keeping the number of HREM section images as low as possible to focus on the regions identified by µMRI and facilitate quick data processing and analysis.

Individually, µMRI and HREM suffer from limitations. The sequential pipeline approach builds on the strength of each and results in a powerful technology allowing for quick and precise morphological phenotype screening of the embryos of transgenic and randomly generated mutants.