SEARCH

SEARCH BY CITATION

Keywords:

  • embryology;
  • models;
  • optical imaging;
  • three-dimensional;
  • tomography

Abstract

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References

Optical projection tomography (OPT) is a new technique for three-dimensional (3D) imaging of small biological tissues. It is particularly useful for reconstructing vertebrate embryos and for examining the 3D anatomy of developing organs. The advantages of this technique over previous methods will be explained: in particular, its ability to image at a higher resolution than magnetic resonance imaging (MRI), while at the same time being able to image specimens much larger than those possible using confocal laser-scanning microscopy. Being an optical technique, OPT is also able to take advantage of the many coloured and fluorescent dyes which have been developed for tissue-specific or gene-specific staining. This becomes particularly important for the visualization of the 3D shapes of specific organs and tissues as it allows the computer to automatically determine the outline of the desired structure.


Background – filling in the gap

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References

The study of anatomy is an intrinsically three-dimensional (3D) endeavour. For centuries anatomists have wished to grasp the full 3D complexity of the structures they study, as much for the intrinsic beauty of these shapes, as for the fact that it helps to appreciate how organs function, how they relate to their neighbouring organs and also how their complex shapes are created during development. Until recently, the only techniques available required creating real physical models (for example, injecting substances that would mould to the shape of the organ and remain intact while the biological tissue was dissolved away). The development of two modern technologies during the last few decades therefore represented an important advance in our ability to appreciate the 3D shapes of tissues and organs: (A) technologies such as X-ray CT scanners and magnetic resonance imaging (MRI) were developed to rapidly and easily generate digital data on the internal structure of tissues without the need for cutting them (i.e. non-invasively), although even today the results are most commonly viewed in two dimensions, as virtual sections. (B) Computer graphics technology allowed for the first time the 3D shapes of biological models to be explored, rotated and manipulated without having to resort to building a real physical model. The raw data for such models may come either from non-invasive scanning technologies such as those mentioned above, or from specimens which have been cut into hundreds of serial sections and reconstructed using computer software. While this latter approach can produce very useful reconstructions, it is too time-consuming and labour-intensive for most researchers to perform – the speed and ease of non-destructive scanning technologies (if they can image what we want to see) make them clearly preferable.

While CT and MRI have become widespread and routine techniques within hospitals for examining tissues the size of human organs, only recently have attempts been made to increase their resolution to analyse specimens as small as early vertebrate embryos (Jacobs & Fraser, 1994). These approaches show promise for a number of applications, and in particular it is possible that in the future microscopic MRI will allow us to image living embryos in 3D during development. However, they also suffer a number of drawbacks. As the intended resolution of MRI increases so does the required strength of the magnet used, such that systems designed to image mouse embryos become too expensive for most laboratories. Also, the achievable resolution of MRI is not usually sufficient to identify all the tissues or organs within the embryo. Additionally, since they are not optical techniques neither MRI nor CT scanning can image the distributions of commonly used staining techniques, such as histological stains, standard immunohistochemical protocols, or in-situ hybridization techniques used to visualize the RNA expression patterns of genes.

At the other end of the size scale is a well-established scanning technology which is indeed optical, but which has not been developed to image anatomy. Confocal laser-scanning microscopy is very efficient at generating clear, 3D images of specimens which have been fluorescently labelled (Potter et al. 1996). However, it is typically used on specimens up to only a few hundred micrometres in thickness, and usually much less than that. As such it is often used for exploring the subcellular distributions of fluorochromes within small groups of cells. An ‘imaging gap’ has therefore been left between confocal and MRI, and unfortunately for developmental anatomists most vertebrate embryos fall precisely in that gap – too large for confocal imaging, and too small for MRI. We have developed a new optical imaging technique, OPT microscopy (Sharpe et al. 2002), which fills this gap (it is ideally suited for specimens between 0.5 mm and 10 mm) and in addition displays several advantages over both of these techniques which will be described in the following sections.

How it works –‘pinpointing’ vs. ‘seeing the whole picture’

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References

Confocal microscopy generates a 3D image of a specimen by focusing a laser beam to a small point within the tissue, and detecting fluorescent light which emerges from that same point. The approach is often called ‘optical sectioning’, as it involves focusing the laser to a specific depth within the tissue (typically sitting on a standard glass slide) and scanning the beam within a horizontal plane, recreating a virtual 2D section. Extending this approach to full 3D imaging simply involves moving the specimen up or down (by raising or lowering the microscope stage), thereby generating a series of optical sections through the tissue. With current confocal microscopes this approach appears unable to generate high-quality images at a depth greater than about half a millimetre.

OPT microscopy uses a very different approach. Rather than reducing the depth-of-focus as much as possible so as to pinpoint only a precise depth within the tissue (as in confocal microscopy), the OPT scanner tries to maximize its depth-of-focus. This results in images with a view right through the whole specimen. These raw data therefore do not explicitly contain information about depth. Instead the technique relies on taking images of the specimen from many different angles, and then using computer software to recalculate the original 3D information. It is in fact an optical equivalent of X-ray CT scanning which uses a very similar principle, the main difference being that whereas the detectors in a CT scanner record a quantitative shadow of the object, OPT uses image-forming optics to create a focused image on a charged coupled device (CCD) camera chip. For the high-resolution results from our experiments we typically bathe the specimen in an organic solvent during image-capture (a mixture of benzyl alcohol and benzyl benzoate) to reduce its opacity, and take 400 images for one complete 360 degree revolution.

Good old-fashioned light – seeing the wood for the trees

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References

One of the advantages that MRI possesses over optical projection tomography (OPT) is that it can ‘see’ through specimens that are far too opaque for optical imaging. It works by aligning all the hydrogen atoms of the specimen within a strong magnetic field, subjecting them to pulses of radiowaves and then detecting the 3D distribution of further radiowaves that are emitted as a response to the excitation (Jacobs & Fraser, 1994). For this reason it is ideal for specimens as large as humans, which are far too opaque for most optical approaches. However, this also creates the disadvantage that the decades-worth of optical stains and dyes that have been developed for studying tissues cannot be used. The same disadvantage applies to X-ray CT scanning, whose rays are too powerful to be disturbed by coloured precipitates. OPT microscopy, however, based on rays in the visible part of the spectrum, can in principle image the distribution of any of these stains. As I will describe here, this is more important than just the ability to map distributions of molecules within the tissue – it is essential if we want a rapid method to see the 3D shapes of particular structures.

If presented with an image of a 2D microscope slide that has been labelled with a standard histological stain, it is extremely difficult for a computer program to identify an arbitrary organ of interest. Given that each organ appears as a collection of light and dark regions within the section, in most cases it is impossible to define the location and shape of the organ based simply on a measure of how dark or light it is, or its colour. By contrast, an anatomist (using far more complex pattern-recognition abilities) will have no trouble picking out, for example the pancreas, as long as the image displays enough contrast between adjacent tissues. MRI is therefore an excellent technique if the resulting virtual sections are to be analysed only by humans, as it displays good contrast between different tissue within a mouse embryo (over 11.5dpc) (Dhenain et al. 2001).

However, as soon as we move to 3D representations of tissues, good contrast is not enough. If we wished to create a 3D surface from an unstained embryo (scanned either by MRI or OPT) by including all regions above or below a certain intensity (a process known as ‘thresholding’) almost all of the organs of the embryo would be included in the model as a single complex shape – we would be unable to see the wood for the trees. We could generate a good 3D shape of the surface of the whole embryo (which can also be useful for certain experiments) but not specific parts within it. To create meaningful 3D surfaces the computer must somehow be able to pick-out the internal structures we wish to study. This is where tissue-specific labelling becomes critical. Suitable labelling techniques have been around for many years, particularly in the form of techniques that highlight the distribution of a specific protein, or the mRNA of a specific gene (Hammond et al. 1998; Sharpe et al. 2002). For example, whole-mount immunohistochemistry can reliably highlight which tissues a protein is expressed in: an antibody is raised against the protein of interest and used to localize a fluorochrome to those tissues where the protein is found.

Using such an approach to define where an organ is located is of limited importance if we are examining a 2D section because, as already mentioned, an anatomist can usually find the organ of interest quite easily using a general histological stain. However, the situation changes dramatically when we want to see the shape of this organ in 3D. Since the computer must be able to detect the 3D outline of the organ, the only alternative to specific labelling techniques is to manually examine every section (whether real or virtual) and instruct the computer where the organ is located. This process (sometimes called ‘painting’) is extremely time-consuming as the number of sections necessary usually runs into the hundreds. This is the approach used in two mouse atlas projects currently underway – one in the MRC Human Genetics Unit in Edinburgh which is based on serial sections (Davidson & Baldock, 2001), and the other in the BioImaging Center at Caltech which is based on MRI (Dhenain et al. 2001). There are at least two reasons why this labour-intensive painting process can be employed in these projects: (1) each embryo will have hundreds of tissues labelled – far more than can be achieved by whole-mount staining techniques, and (2) each embryo will become a standard model for developmental biology and therefore justifies the amount of effort involved. However, this manual painting process is far too laborious for routine inspection of tissue shapes.

This illustrates the advantage of a 3D scanning technique that uses visible light. While new compounds are being developed that are detectable in an MRI scanner, this research is still in progress (Louie et al. 2000). By contrast, OPT can create 3D images of fluorescent signals that are already commonplace in the research of many biologists. In particular it means we can easily image multiple independent signals within the same specimen. Figure 1 shows the results from an OPT scan of a 10-day-old mouse embryo which was stained for the distribution of two proteins. Three different wavelength channels were used: one to observe the autofluorescence of the specimen, which indicates the general anatomy/histology of the whole embryo (shown in red), a second to observe the expression of the protein HNF3b, which is expressed in the endoderm and the floorplate of the neural tube (shown in blue), and a third to observe the expression of the protein neurofilament, which is expressed in the developing nerves (shown in green). In the surface model (on the right-hand side of the figure) we can visualize the 3D shapes of the alimentary canal, floorplate and nerves only because they were specifically labelled.

image

Figure 1. OPT scan of a wildtype 10-day-old mouse embryo. The embryo was whole-mount stained for the expression patterns of two proteins using fluorescently labelled antibodies. In blue is the signal detected for antibodies against the HNF3b protein, and in green is the signal for neurofilament – a protein that is expressed in nerve cells. The red signal represents non-specific autofluorescence from the tissue, which highlights the overall anatomy of the embryo. This same data are displayed in two forms: on the left is a ‘virtual section’ through the reconstruction, and on the right is a 3D surface model of the same embryo.

Download figure to PowerPoint

Not only fluorescence – the advantage of seeing the darkness

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References

Because the principle of OPT is different from confocal microscopy, i.e. focusing through the specimen and rotating it, rather than pinpointing regions of the tissue, it displays another advantage over confocal microscopy. Namely, that it can create 3D reconstructions of signals that are not fluorescent. Again, the chief advantage of this relates to the history of staining techniques – there are still some types of assay for which a reliable fluorescent detection method has not been found. The best example is whole-mount detection of the expression of mRNAs. While the detection of proteins using fluorescently labelled antibodies works very well, the detection of RNAs requires such an amplification of the signal that the most reliable method today still involves the use of alkaline phosphatase to catalyse the conversion of BCIP/NBT to a purple precipitate. Another example of a colour-reaction is the conversion of X-gal to a blue precipitate by the commonly used LacZ reporter gene.

OPT microscopy can be operated in two different modes. The first, already described above, measures the emission of light from a specimen (e.g. from fluorescent compounds). The second mode involves shining light through the specimen and detecting how much is absorbed (e.g. by dark or coloured stains). In the example shown in Fig. 2, the head of a 13-day-old transgenic mouse embryo was stained with X-gal to examine the distribution of LacZ-expressing cells within the developing brain (A. Wilkie et al. unpubl. obs.). During the OPT scan, white light was shone through the head and was partially absorbed by the blue colour of the X-gal precipitate in the expressing cells. The 3D localization of this absorption pattern enabled the OPT software to recreate the positions of these LacZ-positive cells, while the weaker absorption caused by the unstained tissue provided a 3D image of the developing brain. In Fig. 2 the white surface represents the shape of the ventricles – the unstained tissue itself is above this level. The tissue was represented this way so that the LacZ cells can be seen (blue shapes) – if the tissue had been visualized as a solid structure it would not be possible to see the LacZ cells as they are embedded within it. Another example of transmission imaging is shown in Fig. 3. In this case a TS21 mouse embryo was labelled using Alcian Blue – a stain which highlights regions of cartilage, thereby illustrating the shape of the developing skeleton. Although the stain is blue, the OPT scan was monochromatic, so the colours seen in the computer rendering merely represent the optical density of the tissue.

image

Figure 2. Four different views of an OPT scan of LacZ-expressing cells within the brain of a 13-day-old mouse embryo. The blue shapes show where the labelled cells were located. The white surface represents the lowest part of the brain tissue, i.e. the surface of the ventricles. If the tissue itself has been shown as a solid surface, then the LacZ cells would not be visible, as they are embedded within this tissue. Note that although the LacZ cells are shown in blue in this computer representation, the original imaging of this brain was performed in black-and-white. As such, the positive cells were distinguished simply by being darker than the surrounding tissue. This means that some other dark tissue in the brain (for example where blood has developed) is also visible as blue shapes in the images shown here (for example the blue shapes at the bottom of the fourth panel).

Download figure to PowerPoint

image

Figure 3. False-colour computer-generated images of a TS21 mouse embryo which was stained with alcian blue and then scanned by OPT. The images illustrate another advantage of 3D data – that the tissue can be displayed with varying degrees of opacity, thereby allowing one to see more or fewer of the internal structures. (Staining and volume rendering of this embryo by Seth Ruffins at Caltech.)

Download figure to PowerPoint

An extra dimension in understanding

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References

One of the applications for OPT is analysis of altered morphology in mutant mice. The advantage of performing this analysis in 3D rather than 2D sections is clearly illustrated in Fig. 1. The positions of the nerves within the section (left) are easily visualized as small green regions, so if a mutant lost most of its nerves this would be easily noted. However, if the mutant phenotype displayed incorrect innervation, for example if nerves from the wrong axial position were growing into the limb buds, that would be very difficult to discover from the sections because the connectivity of the nerves is virtually impossible to work out. By contrast, even subtle changes in wiring could easily be observed from the 3D model (right). This type of advantage has already been seen with recent analysis of the Bagpipe mutant. Although certain aspects of its phenotype have been known for a number of years (for example loss of the spleen), until we performed OPT scans on the internal organs of these Bagpipe embryos it was not known that the stomach displays a series of abnormal ridges and grooves (Sharpe et al. 2002). This phenotype, quite dramatic when viewed by OPT, was simply too subtle to be noticed by more conventional techniques.

The developing brain in Fig. 2 is another example where seeing the 3D arrangement of the LacZ-positive cells in relation to the ventricles gave an understanding of their distribution that was not possible from sections alone. The previous analysis of these specimens, which had been done by cutting paraffin wax sections, produced conclusions which had to be revised once the OPT models were available. On a more general level, in our experience every time we have viewed an OPT-generated 3D model of embryonic structures we have learned something new about them. Given the intrinsically 3D nature of the structures that developmental biologists work with every day it is ironic how unfamiliar we are with seeing their true shapes. We hope that OPT microscopy will change this unfortunate situation, and maybe even encourage a brief new phase of descriptive embryology – this time based firmly in three dimensions.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References

I wish to thank the following colleagues for helping me develop this technique and providing specimens to test it with: Paul Perry, Bill Hill, Richard Baldock, Ulf Ahlgren, Allyson Ross, Jacob Hecksher-Sørensen, Rob Bryson-Richardson, Anemieke Ijpenberg and invaluable support from Duncan Davidson and Nick Hastie.

References

  1. Top of page
  2. Abstract
  3. Background – filling in the gap
  4. How it works –‘pinpointing’ vs. ‘seeing the whole picture’
  5. Good old-fashioned light – seeing the wood for the trees
  6. Not only fluorescence – the advantage of seeing the darkness
  7. An extra dimension in understanding
  8. Acknowledgements
  9. References
  • Davidson D, Baldock R (2001) Bioinformatics beyond sequence: mapping gene function in the embryo. Nature Rev. Genet. 2, 409417.
  • Dhenain S, Ruffins SW, Jacobs RE (2001) Three-dimensional digital mouse atlas using high-resolution MRI. Dev. Biol. 232, 458470.
  • Hammond KL, Hanson IM, Brown AG, Lettice LA, Hill RE (1998) Mammalian and Drosophila dachsund genes are related to the Ski proto-oncogene and are expressed in eye and limb. Mech. Dev. 74, 121131.
  • Jacobs RE, Fraser SE (1994) Magnetic resonance microscopy of embryonic cell lineages and movements. Science 263, 681684.
  • Louie AY, Huber MM, Ahrens ET, Rothbacher U, Moats R, Jacobs RE, et al. (2000) In vivo visualization of gene expression using magnetic resonance imaging. Nature Biotechn. 18, 321325.
  • Potter SM, Fraser SE, Pine J (1996) The greatly reduced photodamage of 2-photon microscopy enables extended 3-dimensional time-lapse imaging of living neurons. Scanning 18, 147.
  • Sharpe J, Ahlgren U, Perry P, Hill B, Ross A, Hecksher-Sorensen J, et al. (2002) Optical projection tomography as a tool for 3D microscopy and gene expression studies. Science 296, 541545.