2.1.

Staining and Embedding Procedure

Murine CNS tissues from previously dissected laboratory mice (75d old, C57N/Bl6 mice, sacrificed for a different biomedical research project) were used to showcase and compare the staining agents of interest here. The effective uptake of ${\mathrm{OsO}}_4$ by various retinal layers was already demonstrated in a preceding high resolution XPCT study.³⁰ Besides fully intact eyes, tissue from murine cerebellum and corpus callosum was collected to investigate the binding behavior of the different staining agents, namely ${\mathrm{OsO}}_4$, Lugol’s iodine solution (LIS), and NdAc.

The mice were perfused or immersion-fixed with a K+S solution. After several washing steps in 0.1 M phosphate buffer (PB) ($3\times 15\min$), the tissue ($\mathrm{diameter}\sim 1.5\mathrm{mm}$) was post-fixated with 2% ${\mathrm{OsO}}_4$ at 4°C (4 h). To this end, 4% ${\mathrm{OsO}}_4$ stock solution was mixed in a 1:1 ratio with 0.2M PB. In the next step, the tissue was repeatedly washed ($3\times 15\min$) in the same buffer and dehydrated with increasing concentration of acetone (1 h) and embedded by infiltration with 2:1 acetone/EPON (1.5 h), 1:1 acetone (1.5 h), 1:2 acetone/EPON (overnight), and pure EPON (4 h).³¹ Finally, the samples were mounted in polyimide (Kapton) tubes and polymerized for 24 h at 60°C. Staining with ${\mathrm{OsO}}_4$, LIS, and NdAc was conducted in the same way. The staining protocols for ${\mathrm{OsO}}_4$ and LIS are already well established in X-ray tomography and proved to enhance contrast by binding to different regions of biological tissue.^21,32

Since Nd and U belong to the same side group of the periodic table, similar chemical properties with regard to binding and clustering at different regions is expected and can indeed be observed, see for example Fig. 1(c). In buffer, Nd is present as a trivalent cation that accumulates preferentially at anionic groups, in particular phosphate groups in nucleic acids, phosphate, and carboxyl groups, such as gangliosides, mostly present on cell surfaces.³³ NdAc is applied in aqueous solution (4%) to give sufficient contrast, which could be further enhanced by a higher NdAc concentration or, e.g., a replacement of water by a saturated methanolic solution.³⁴ In contrast to UAc, NdAc is neither radioactive nor highly toxic, significantly facilitating the handling and staining procedure.

Fig. 1

(a) Schematic overview of the parallel beam setup at the GINIX endstation with an attenuator in and KB-mirrors and X-ray waveguide out of the beam path (right). The propagated exit wave of a given sample projection impinges on the detector at a distance $x_{12}$ behind the sample, chosen to reach the holographic regime (left). (b) Laboratory nano-CT imaging setup with opening angle $\theta$, a spatial coherence length ${\xi }_{\perp }$. The ratio of source-to-sample ($x_{01}$) and sample-to-detector ($x_{12}$) distance (left) defines the magnification $[(x_{01}+x_{12})/x_{01}]$. Photo of laboratory setup (right): sample holder (green), beam (red), and detector area (blue). (c) 2D chemical structure of NdAc. A trivalent cation is coordinated by three acetate molecules that are formed by combination of acetic acid with a base. All lengths are in mm.

2.2.

Laboratory Nano-CT Imaging

While synchrotron imaging is still regarded as the gold standard in XPCT, state-of-the-art laboratory sources can provide sufficient flux and coherence for in-house (phase-contrast) tomography with voxel sizes in the range of a few microns down to the sub-micron range. Recent results have opened up a promising approach to study biological tissues in the laboratory^18,35,36 and to translate 3D virtual histology from synchrotron to compact laboratory sources, important for pre-clinical and clinical research.

While the tissue samples have been scanned at several different custom built in-house CT instruments, we here present results obtained at the commercially available EasyTom nano-CT system (RX Solutions, France). The system allows for switching detectors and sources and flexible adjustment of geometry and source parameters well suited to optimize the setting for a specific application. In the present case, the nano-CT source was used with a LaB6 cathode and a transmission anode with an adjustable spot size (0.4 to $1.5\mu \mathrm{m}$), a CCD camera (Gadox scintillator with free fiber optics coupling, $9\mu \mathrm{m}$ pixel pitch, and $2016\times 1344\mathrm{pixel}$ after $2\times 2$ binning), and a custom-made sample holder. High geometric magnification was selected by moving the sample close to the source spot Fig. 1(b). $N_{\mathrm{prj}}=1568$ images were recorded over 360 deg with a tube voltage of 80 keV and a polychromatic spectrum. Accumulation of several images at the same position was used to improve the signal-to-noise ratio (SNR). Before the scan, the reference images at selected angles were recorded for subsequent tracking and correction of spot size movement over the duration of the scan. Flat field images were recorded inside the EPON material to receive normalized intensities in the projection images by correction for low-frequency effects.³⁷ The exact settings are given in Table 1.

Table 1

Parameters of tomographic scans for stain comparison at laboratory source.

After the scan was completed, the reference images were used to account for source spot movement, and the rotation axis was corrected. Finally, ring filter correction and phase retrieval algorithm were applied before the actual tomographic reconstruction. For phase reconstruction, a Paganin-based phase retrieval algorithm was used and the according parameter ($\delta /\beta$) was set manually.³⁸ Fiji³⁹ and MATLAB (R2020a) were used for further analysis and NVIDIA IndeX⁴⁰ for rendering of the volumes. Video 1 shows an entire rendered optical nerve with retina scanned at the laboratory setup (Avizo, Thermo Fisher Scientific, Germany).

2.3.

Synchrotron-Based X-Ray Tomography

SR sources provide users with highly brilliant and coherent X-ray beams that allow one to exploit quantitative phase contrast on multiple scales and correspondingly different geometries. In this experiment, the parallel beam tomography setup⁴¹ of the GINIX endstation⁴² at the P10 beamline of the PETRA III storage ring (DESY, Hamburg) was used. Due to the high number of photons, a continuous rotation and short exposure time is sufficient to fill the dynamic range of a 16-bit detector, resulting in reduced moving artifacts and a short overall scan time ($\approx 2\min$). For the present measurements, single-distance tomograms with 3000 projections over 360 deg were recorded. The monochromaticity ($\mathrm{Si}(111)$ channel-cut monochromator) allows the tissue to be studied at a well defined photon energy instead of the broad bandpass unavoidable for in-house CT. The sample’s optical indices and imaging parameters fall within the regime where CTF-based phase retrieval is suitable to solve the phase problem.⁴³

For the setup, the sCMOS camera pco.edge 5.5 (PCO, Germany) with a 100 Hz maximum frame rate, 2560 × 2160 pixel, a rolling shutter, and a fast scan mode was mounted on an high resolution detection system (Optique Peter, France) with a 50 mm-thick LuAG:Ce scintillator and a $10\times$ magnifying microscope objective.⁴¹ This resulted in an effective pixel size of $0.65\mu \mathrm{m}$ and a total exposure time of about 75 s. Focusing optics such as KB-mirrors or waveguides as well as the fastshutter were moved out of the beam path, whereas the beam size was adjusted by an upstream slit system resulting in a beam size of about $2\mathrm{mm}\times 2\mathrm{mm}$ [Fig. 1(a)].

With a sample-to-detector distance (SDD) of 149 mm, the holographic regime was reached with a Fresnel number of

Tomographic reconstruction was carried out after flat- and dark field correction with the iradon function implemented in MATLAB using a Ram-Lak filter for backprojection. To avoid ring artifacts, hot pixel in the projections was removed by replacing them with the mean of adjacent pixel. The Fresnel number was verified and if needed refined by inspecting the radially averaged PSF of a hologram to the expected CTF and according re-scaling. In addition, a wavelet-based filtering approach was applied to further mitigate ring artifacts occurring during the reconstruction.⁴⁶ For post processing and visualization, the Fiji software was used.

3.1.

Laboratory Comparison

3.1.1.

Murine eye imaging

We first present in-house scans of EPON embedded murine eyes, stained either with NdAc or ${\mathrm{OsO}}_4$ as well as an unstained reference sample, see Fig. 2(a). Already based on visual inspection of the projection images, which were recorded with the exact same settings (Table 1), clear differences between the samples can be observed. While the unstained tissue shows a general lack of contrast, which is reflected by a very low SNR in the reconstructed images, different regions of the retina are contrasted and emphasized in the stained preparations. The projections of the NdAc treated retina show mainly the surfaces/interfaces of the tissue to be contrasted, whereas ${\mathrm{OsO}}_4$ seems to reach a higher and more bulk-like penetration and binding, resulting in a more uniform and stronger attenuation in all regions. A qualitative analysis of the intensity profiles across the retinal layers in ${\mathrm{OsO}}_4$ and NdAc stained layers indicates larger deviations of attenuation values depending on the according layer, thus suggesting a higher specificity of the NdAc stain. In addition, different retinal layers show higher absorption properties. While, e.g., in the ${\mathrm{OsO}}_4$ stained sample mostly the choroid layer is strongly targeted, the NdAc shows a high binding affinity to the sclera, the internal limiting membrane as well as rods and cones in the retina. In addition, cells in the Ganglion cell layer are highlighted and their distribution can be investigated. We also note that the NdAc stained region seems to be slightly compressed in comparison to the ${\mathrm{OsO}}_4$ stained sample ($\approx 60\mu \mathrm{m}$ thinner), which might be attributed to the selection of the compared slices as well as to sample preparation artifacts or differences in specimens.

Fig. 2

Comparison between NdAc (left), ${\mathrm{OsO}}_4$ (center), and unstained (right) entire murine eye samples by in-house nano-CT system. (a) First row shows a projection image of each dataset. The second row depicts a selected slice from the $xy$-plane (maximum intensity projections over five slices) and the third row from the $yz$-plane. In (b) and (c), smoothed line profiles from a similar region in the ${\mathrm{OsO}}_4$ and NdAc sample are plotted. Each colored region represents a layer of the murine eye. Yellow, sclera; green, choroid; light blue, rods and cones layer; dark blue, nuclear/plexiform layers; pink, ganglion cell layer. Scan settings and reconstruction parameters are given in Table 1. Scale bars: $200\mu \mathrm{m}$.

3.1.2.

Murine brain imaging

Murine corpus callosum and cerebellum were also imaged in the laboratory setup to investigate the binding behavior of NdAc in different neuronal tissues (Fig. 3; Table 1). It can be observed that in both cases, surfaces and interfaces are pronounced, possibly because these interfaces are defined by well exposed cell surfaces with anionic phospholipids. Also, vasculature seems to be affected by the stain and the whole vascular network becomes easily distinguishable from the surrounding tissue, which opens up a promising approach for rendering and segmentation of the vascular tree. In the cerebellum, gray matter is also clearly separable from white matter. Cell layers on the other hand could not be clearly be resolved. This could be explained based on insufficient sampling and resolution, or—more likely—by a lack of contrast, which could either result from a low penetration depth of the stain, the compactness, and higher tissue density, or to the lower binding of Nd in these regions.

Fig. 3

(a) Corpus callosum and (b) cerebellum stained with NdAc shown in $xy$-plane (top) and $xz$-plane (bottom), both as maximum intensity projections over five slices, recorded with the laboratory setup. Scan settings and reconstruction parameters are listed in Table 1.

3.2.

Qualitative Comparison of Different Staining Agents Using Synchrotron Phase-Contrast CT

Synchrotron data were collected from a NdAc, ${\mathrm{OsO}}_4$, Lugol’s solution stained retina, and compared with an unstained sample at 13.8, 16, and 20 keV. In Fig. 4, reconstructed slices are presented for all samples scanned at 16 keV since this photon energy yielded the most comparable results.

Fig. 4

Comparison between NdAc, ${\mathrm{OsO}}_4$, LIS stained and unstained entire murine eye samples scanned in parallel beam geometry at 16 keV at the synchrotron facility (SDD: 149 mm). (a) A selected slice from the xy-plane (scale bar: $200\mu \mathrm{m}$) and (b) a slice from the $yz$-plane with a zoomed section (scale bar: $50\mu \mathrm{m}$). Scan settings and reconstruction parameters are given in Table 2.

Table 2

Parameters of synchrotron measurements of different stains with a parallel beam setup.

Here, our previous assumptions following the qualitative assessment of the laboratory scans can be confirmed, however, now at much increased resolution and contrast. NdAc seems to significantly enhance contrast of cones and rods as well as the sclera and the Ganglion cell layer while ${\mathrm{OsO}}_4$ and LIS target the choroid. While cones and rods also seem to be targeted by the stains, ${\mathrm{OsO}}_4$ lacks any specificity and increases the attenuation in the sample uniformly, with the heavy osmium atoms causing a strong attenuation of photons in the tissue. LIS on the other hand is very favorably binding to the choroid, decreasing the number of photons passing through the sample, thus decreasing the SNR, while other layers of interest in the retina seem to only experience a minor increase in contrast. It is worth noticing that ${\mathrm{OsO}}_4$, which is known to strongly bind to lipids, yields strong contrast for adipose tissue and eye muscles right behind the eyeball, allowing for an easy segmentation and detailed investigation of these structures [Fig. 5(b)].

Fig. 5

Rendering with NVIDIA IndeX. (a) A part of a vascular tree extracted from a NdAc stained murine corpus callosum [see Fig. 3(a)]. Scale bar: $40\mu \mathrm{m}$. (b) An ${\mathrm{OsO}}_4$ stained rendered eye sample, where the adipose tissue and eye muscles are clearly visible (green), whereas the optical nerve is indicated and culminates into the retinal layer (turquoise). (c) An NdAc stained rendered eye sample where the sclera, coating the optic nerve, is clearly visible and extends into the grid-like, stiff collagenous structure. (b) and (c) were recorded at the laboratory nano-CT source. (d) An illustration of the eye structure and (e) The retinal layers of the ${\mathrm{OsO}}_4$ stained eye [same sample as in (b)].

3.3.

Quantification of Staining Results

Tomography with monochromatic SR allows to draw quantitative conclusions on the composition when modeling the absorption coefficients and/or the reconstructed phase shifts based on atomic form factors. Since the phase retrieval scheme used here yields images of coupled phase and absorption contrast based on its homogeneous object approximation, it is per se not straightforward to extract the local concentration of the stain, i.e., a spatially varying stochiometry. However, in propagation based XPCT, the contrast of the low spatial frequencies is dominated by absorption. Therefore, we can artificially generate a “pure” absorption image by local averaging the projections such that the phase effect cancels out. This is done only for the purpose of quantifying the local concentration in a coarse-grained sense, see the corresponding workflow shown in Fig. 6(c). The reconstructed absorption coefficients in the reconstruction volume are then compared between a NdAc stained sample and an unstained sample, see Fig. 6(b), scanned both at 13.8 and 16 keV. To this end, the projections are filtered, e.g., with a median filter ($8\times 8\mathrm{pixel}$), allowing one to take only the attenuation into consideration. A comparable region of interest involving the retinal layers with $400\times 400\mathrm{pixel}$ (16 keV) is then cropped from a reconstructed slice. Histograms of the gray values are shown in Fig. 6(d). By subtraction of the attenuation peaks, we obtain the additional attenuation coefficients (or more precisely the histogram thereof) attributed to NdAc in units ($1/vx$), allowing us to effectively isolate the staining component from the tissue and to infer its atomic density based on the known atomic form factor ($f^{\prime \prime }$ component).

Fig. 6

Quantitative analysis of the NdAc staining from synchrotron measurements. (a) Reconstructed slice (left) with and (right) without phase retrieval, demonstrating the benefit of phase retrieval for the image quality and quantitative contrast. The example shows NdAc stained retinal layers recorded with the synchrotron setup. (b) Qualitative comparison between NdAc stained (left) and unstained retina (right) at 13.8 keV. Scale bar: $100\mu \mathrm{m}$. (c) Illustration of the workflow associated with suppressing the phase effects by local averaging of the projection (left) to create an absorption-dominated image (center). The line plot (right) shows the corresponding reduction of edge enhancement. (d) Attenuation per voxel shown for similar region of NdAc, ${\mathrm{OsO}}_4$, and unstained samples in histograms for 13.8 and 16 keV. The red lines mark the attenuation value where the number of counts peaks. By subtraction from the unstained peak, the attenuation of NdAc and ${\mathrm{OsO}}_4$ is determined.

The refractive index

Substituting Eq. (4) into Eq. (3), we obtain an atomic mass density of

Since we know $\lambda$ ($\lambda [\AA ]=12.938/\mathrm{E}[\mathrm{keV}]$), the number density and subsequently the mass density ($\rho ={\rho }_{a}A/N_A$) with the known atomic mass $A$ and the Avogadro constant $N_A$ can be calculated. The atomic scattering factors $f^{\prime }$ and $f^{\prime }$ for uranium, osmium, and NdAc are plotted in Fig. 7, where the absorption edges ($L$, $M$) are indicated. To investigate their optical properties in regard to the relationship between absorption and refraction, the $\delta /\beta$ ratio is calculated as⁴⁷

Fig. 7

Plots of atomic scattering factors $f^{\prime }$ and $f^{\prime \prime }$ for uranium, neodymium, and osmium (with denoted M- and L-edges) and their energy-dependent $\delta /\beta$ ratios [Eq. (6)].

From the plots, a significantly higher ratio can be inferred for neodymium at energies above 17 keV and between 2.5 and 6 keV, where therefore an advantage of neodymium can be expected for phase contrast imaging compared to uranium.

At 13.8 keV, the difference of the attenuation peaks was determined as $1.8250·{10}^{-4}(1/\mathrm{vox})$. Substituting this into Eq. (5), an approximate number density for NdAc of $0.0117(\mathrm{Nd}/{\mathrm{nm}}^3)$ is obtained. To check consistency, the procedure was repeated for samples scanned at 16 keV, yielding a number density of $0.0105(\mathrm{Nd}/{\mathrm{nm}}^3)$. In contrast, ${\mathrm{OsO}}_4$ shows a difference in attenuation peaks from the unstained sample of about $6.2·{10}^{-4}(1/\mathrm{vox})$ resulting in an estimated number density of $0.0249(\mathrm{Nd}/{\mathrm{nm}}^3)$. Number densities and estimated free distances between atoms are listed in Table 3.

Table 3

Number densities and estimated mean distance between Nd atoms for different energies.

4.1.

Animal Welfare

Mice were housed in the mouse facility of the Max Planck Institute of Experimental Medicine with controlled ventilation (inward airflow, exhaust to the outside), temperature ($21\pm 2^{°}\mathrm{C}$) and controlled rel. humidity ($55\%\pm 10\%$) in individually ventilated cages. Mice were group-housed 3 to 5 mice per cage in a 12-h dark/light cycle with ad libitum access to food and water. For the procedure of sacrificing vertebrates for subsequent preparation of tissue, all regulations given in the German animal welfare law (TierSchG §4) are followed.