Optical Projection Tomography as a Tool for 3D Microscopy and Gene Expression Studies James Sharpe, Ulf Ahlgren, Paul Perry, Bill Hill, Allyson Ross, Jacob Hecksher-Sørensen, Richard Baldock, and Duncan Davidson

Technical information about OPT microscopy

If the projection data contained within each image represented exact line integrals through the specimen then the first image would contain the same information as the image taken after a 180 degree rotation (because true projections taken in exactly opposite directions have the same overall orientation - illustrated by the double arrowhead in Fig. 2E). As a result of this, some projection tomography techniques require only a 180 degree rotation to complete a full reconstruction. In our experiments however a different approach was taken. The use of image-forming optics (necessary to achieve the desired magnification) means that objects are only in focus over a finite range, and in some cases this will not be sufficient to include the entire specimen. We overcame this problem by carefully adjusting the position of the specimen relative to the focal region. Instead of placing the axis of rotation exactly on the focal plane (Fig. 2A) the position of the axis was adjusted, such that only the front half of the specimen was in focus (Fig. 2B). This ensures that every part of the specimen is imaged in focus through a full 180 degree rotation. Since each set of projection data now also contains low-resolution information about the "out-of-focus" half of the specimen the algorithm can be altered to take this into account, however in our experiments this did not significantly improve the resulting reconstruction.

To determine whether the data collected approximated parallel line integrals through the specimen, we measured the angles of lightpaths from a typical imaging experiment. Using a 0.5x objective and a magnification of 4, light from a single point on the focal plane is captured by the objective lens within a 10mm diameter circle (cd in Fig. 2C). With a working distance of 135mm this gives the angle a value of 4.2 degrees. This gave a sufficient depth-of-focus to image a specimen which was 4mm across. The intensity recorded at a single pixel approximates a line integral (diagonal black line in Fig. 2C) whose position is determined as the centre of the sampling cone for that pixel (blue lines in Fig.2C). Measuring the angle of these lines showed that the angle of a projection at the outer edge of a 4mm specimen relative to the optical axis was only 0.3 degrees, and therefore an assumption of parallel projections is a reasonable approximation.

As a preliminary test to explore how well the results of a reconstruction compare with the original specimen, we used the principle of a "phantom" specimen (as commonly used in MRI), ie. a artificially-constructed specimen for which the distribution of optical densities is known. Three cylinders of agarose with different concentrations of Chinese ink were embedded within a larger cylinder of the same agarose concentration (Fig. 2D). A reconstructed section through the phantom showed that the calculated values were evenly distributed (ie. the values did not significantly fade towards the edge or the centre of the cylinders - so-called "dishing" effects). The ink in the 3 cylinders was at a concentration of 1x, 2x and 3x, and the reconstructed values approximated a linear response to this (relative values of 50, 99 and 147) suggesting that the technique may be used for quantitative measurements of dye concentrations.

Another approach for testing the validity of a reconstruction is to examine the same specimen using a second method and compare the two. Therefore we explored whether it is possible to obtain good microtome sections from specimens after they have been imaged by OPT microscopy. After scanning, a E11.5 embryo was washed back into methanol and embedded in paraffin wax for sectioning. The sections were then mounted and stained with heamatoxilin and eosin. Comparison between panels F and G from Fig. 2, show that the OPT section is faithful to the original tissue, interestingly showing a similar distribution of grey-values to the H&E staining. This comparison also highlights one of the problems associated with cutting real sections - that the geometry of the tissue has been distorted. The choice of 400 angular positions for imaging was made by starting with a smaller number (180) and increasing until a suitable level of resolution was obtained.

Materials and Methods
Specimen staining

In-situ hybridisations were carried out as previously described (1). Whole-mount immunohistochemistry was performed as follows. Embryos and dissected organs were fixed in 4% PFA in 0.1M PBS at 4°C for 60-90 minutes. They were washed in TBST (0.05M Tris-HCl pH 7.5, 0.15M NaCl, 0.1% Triton X-100 and 0.1% Sodium Azide) and blocked in 10% serum (derived from the species in which the secondary antibody (Ab) was produced) in TBST (also used for all Ab incubation steps) for 4h at RT. Specimens were subsequently incubated with primary Ab for 6h at RT and thoroughly washed in TBST with multiple changes for 6h at RT. Secondary Ab incubation and washing was carried out as for primary. Primary antibodies used were diluted as follows; Rabbit anti-HNF3 1:4000 (Kindly provided by T.M. Jessell), Rat anti-E-cadherin 1:1000 (Zymed, clone ECCD-2) and FITC conjugated mouse anti-neurofilament 1:400 (Sigma). Secondary Ab's used were: Cy3 anti-Rat 1:400 and anti-Rabbit 1:400 (Jackson), and Alexa 488 anti-Mouse 1:200 (Mol. Probes). In general, primary Ab's were diluted half as much as optimised for on sections and secondary Ab's as used on sections.

Imaging

Specimens were embedded in small blocks (generally cylinders) of 1% low melting point agarose which were incubated overnight in 100% methanol, and then at least 4 hours in Murray's Clear (a 1:2 mixture of Benzyl Alcohol and Benzyl Benzoate, Sigma). These were rotated using a stepper motor with 0.9 degree steps, under a Leica MZ FLIII microscope with a 100W mercury-vapour burner for fluorescence imaging. The following Leica filter sets were used: "GFP1" for autofluorescence (425/60nm excitation, 480nm barrier), "GFP2" for FITC and Alexa 488 (480/40nm excitation, 510nm barrier), and "Green" for Cy3 (546/10nm excitation, 590nm barrier). A 12-bit Xillix camera was used for all imaging, controlled by scripts written for the imaging software IPLab (Scanalytics Inc.) running on a Macintosh G3 computer. The camera had pixels dimensions of 1301 x 1036, but images were always halved in resolution by binning 2 x 2 squares of pixels, resulting in images of 650 x 517 pixels for computer processing. This resulted in pixels generally imaging 5 to 10 microns square of the embryo, depending on the magnification used.

Reconstructions

Projection images were corrected for pixel variation and background variation using in-house software written for the Edinburgh Mouse Atlas Project. This corrected data was then transformed and filtered before back-projection. The transformation from light intensity captured at each pixel to optical density required for the back-projection, was performed by inverting the data and raising to the power of 10, although in some cases lower values of the power function produced better results. These aspects of the technique will require a detailed characterisation of the system for further improvement. The best results for fluorescent data were obtained without any transformation before filtering. The filter used before back-projection was a simple ramp filter. (All software for the filtered back-projection was adapted from code written in 1978 and available from the web-site of M. Slaney - www.slaney.org/malcolm/pubs.html). 2-D sections perpendicular to the axis of rotation were back-projected independently, and assembled into 3-D voxel data. These calculations were performed on unix workstations, and lasted between 2 to 5 hours. The virtual sections shown here were not manipulated apart from a linear re-scaling of intensities. 3-D rendering of organs was performed by calculating iso-surfaces, ie. connected 3-D contours which represent a particular intensity of the antibody staining signal. A gaussian smoothing (0.5 voxel radius) was performed on the data before extracting surfaces. Both the iso-surface calculations and the rendering was performed using Visualisation Toolkit (VTK, www.kitware.com). The filtered back-projection software was adapted from code written in 1978, and currently accessible from

Reference

1. K. L. Hammond, I. M. Hanson, A. G. Brown, L. A. Lettice, R. E. Hill, Mech. Dev. 74, 121 (1998).

Supplemental Figure 1. OPT microscopy. (A) A schematic of the OPT microscopy set-up. The specimen is rotated within a cylinder of agarose while held in position for imaging by a microscope. Light transmitted from the specimen (blue lines) is focused by the lenses onto the camera imaging chip (CIC). The apparatus is adjusted such that light emitted from a section which is perpendicular to the axis of rotation (red ellipse) is focused onto a single row of pixels on the CIC (red line). The section highlighted as a red ellipse in (A) is seen as a red circle in (B). The region of the specimen sampled by a single pixel of the CIC is shown as a double inverted cone shape (blue region). Points far from the focal plane will not appear sharply focused in the image (pale blue shading) while those closer to the plane will be more focused (darker blue shading). While confocal microscopy attempts to minimize the noise from out-of-focus regions by illuminating only those points on the focal plane, in our experiments the sample cones were made as narrow as possible (11) such that the sampled region approximates a narrow cylinder through the specimen. As the depth-of-focus was not always large enough to include the entire specimen, the position of the axis of rotation was adjusted such that only the front half of the specimen was in focus. This ensures that every part of the specimen is imaged in focus during a full 360 degree rotation (11). (C) The sampled regions from adjacent pixels are distributed across the section as an approximation of parallel line integrals. Panels (D to L) illustrate three stages during the back-projection of a single section through the specimen. The first row shows the situation after only one image has been processed. (D) shows the image with the section highlighted as a red line. (E) shows a plan view of the section (red circle) and the orientation of the projection data through the section. (F) shows the filtered projection data from the first image back-projected using the orientation shown in (E). The next row shows the situation after the first 100 images have been processed. The specimen has rotated through 90 degrees at this stage (as seen in G), and the orientations which have been processed include all the angles from 0 to 90 degrees (grey lines in H). (I) shows the accumulated projection data at this stage. The last row shows that after 180 degrees of rotation (J) a full set of orientations have been collected (K) and the section is fully reconstructed. However, in OPT microscopy the rotation is continued for another 180 degrees, to increase the resolution.

Medium version | Full size version

Supplemental Figure 2. Characterisation of OPT microscopy. Panels (A) and (B) show the rotation of a specimen (red circle) which has a small black oval structure towards the its outer edge. This structure is shown in 4 positions during the rotation. In the first case (A) the oval leaves the depth-of-focus twice during the rotation - when it is furthest from the lens and closest to the lens. It is therefore never imaged in focus when horizontal. The orientations for which it is imaged in focus are illustrated as grey lines (mostly vertical). By contrast, when the focal region is adjusted to include only the front half of the specimen (B) although the oval is still only imaged in focus during half of the rotation, it is seen at all orientations during this half (the grey lines now include all orientations from 0 to 180 degrees). Although the oval is out of focus for the other half of the rotation, this information does not affect the quality of the reconstruction. (C) The lightpaths for a typical imaging experiment. The angle of the sampling cone () is 4.2 degrees, while the divergence of a projection (diagonal black line) coming from the edge of a 4mm specimen is 0.3 degrees with respect to the optical axis (horizontal black line). d2 is the distance from the optical axis of a point on the edge of the specimen which is also on the focal plane. cd is the diameter of the circular region which captures the imaged light from that single point. d1 is the distance of the centre of the circular region from the optical axis. Panel (D) shows the Chinese ink phantom used for preliminary characterisation of OPT imaging, and (E) shows a calculated section through the phantom. The intensities of the reconstructed ink cylinders is proportional to the concentrations of the ink used. (F), A virtual section through an unstained E11.5 mouse embryo, and (G), a real paraffin wax section cut through the same specimen.

Medium version | Full size version