The role of cell body density in ruminant retina mechanics assessed by atomic force and Brillouin microscopy

Eyes were removed from ewes 10–15 min post mortem and kept in PBS (Oxoid) on ice. Retinae were extracted 30 min later and fixed in 0.5% PFA (Sigma; in PBS) for 30 min at room temperature. Small pieces of retinal tissue were embedded in OCT (TissueTek Optimal cutting temperature compound, VWR) and cooled down on dry ice. 14 µm thick sections were cut on a cryostat (Leica CM3050) and transferred to super frost objective slides (Thermo Scientific). A minimum of 3 sections per retina from 2 sheep were used for analysis. Bovine eyes were extracted from 2 year-old female bovine and kept on ice for transportation. Retinae were extracted and fixed in 10% formalin for  >48 h at room temperature. Then, 5 µm-thick slices were cut for histologic analysis. A minimum of 4 sections from 2 cows were stained. Sections were rehydrated using Ethanol (Sigma; 100% for 3 min, 95% for 2 min, and 75% for 2 min) and rinsed in water for 5 min. Harris' hematoxylin solution (pfmMedical) was applied for 4 min before slides were washed with water for 5 min. Sections were subsequently stained with alcoholic Y eosin solution (Leica) for 1 min and rinsed in water for 40 s. Dehydration was achieved by washing with Ethanol (75% for 40 s, 95% or 1 min, and 100% for 3 min), followed by applying xylene (Fisher Scientific) for 3 min for clearing. Sections were mounted in DPX mountant (Fisher Scientific) and imaged on a Zeiss Axioskop with a QImaging MicroPublisher 5.0 RTV camera attached to it.

For AFM measurements, eyes were removed from adult female sheep 10–15 min post mortem and kept in phosphate buffered saline (PBS, Oxoid) on ice. Retinae were extracted within 3 h post mortem. Pieces of retinal tissue were bathed for 5–10 min in ice cold PBS containing MitoTracker Orange (ThermoFisher, 1 µM) [28], washed with fresh PBS and subsequently embedded in 2.5% low melting point agarose (Sigma; 37–40 °C in PBS). Using superglue, the retina-containing agarose block was glued on a vibratome (VT1000 S, Leica Microsystems) plate and placed in the vibratome basin containing ice-cold PBS. 300 µm thick cross sections were cut using one half of a Gillette 7 O' Clock double edged razor blade with a frequency of 75 Hz and a forward speed of ~50 µm s⁻¹. Slices were transferred to BD Cell-Tak-treated (Cell and Tissue Adhesive; BD Biosciences) glass slides, covered with PBS, and used for AFM measurements. For Brillouin microscopy measurements, freshly enucleated bovine eyes were obtained from a local slaughterhouse and kept on ice during the transportation until the starting of the experiment. Retinae were dissected immediately before the measurements, transferred to filter paper, slices cut with a scalpel blade, and mounted upright using a custom-built chamber [28] filled with L15 medium (ThermoFisher). All experiments were completed within 8 h after sacrifice.

14 µm thick sheep retina cryo-sections were collected on super frost objective slides. For visualization of nuclei, sections were treated with DAPI (Sigma Aldrich) for 15 min, washed with PBS, and mounted in Fluoromount Aqueous Mounting Medium (Sigma Aldrich). Fluorescence images were taken on a Leica SP8 confocal laser scanning microscope (40  ×  oil objective, NA  =  1.3). 3 images of different parts of the retina were subsequently processed using FIJI imaging software. 8-bit images were converted into binary images, watershed applied, and a particle analysis (Size  =  1  −  Infinity, Circularity  =  0.00–1.00) was performed for the outer nuclear layer (ONL), inner nuclear layer (INL), and inner plexiform layer (IPL). Three measurements per layer and image were performed, and each layer's relative mean area covered by cell nuclei A_norm was calculated.

Force-distance curves were acquired using a JPK Cellhesion 200 (JPK Instruments AG, Berlin, Germany) setup on an inverted optical microscope (Axio Observer.A1, Carl Zeiss Ltd. Cambridge, UK), and fluorescently labelled retinae imaged simultaneously. Tipless silicon cantilevers (Arrow-TL1, NanoWorld, Neuchatel, Switzerland) with spring constants of ~0.03 N m⁻¹ were custom-modified by attaching polystyrene beads of d  =  10 µm (microParticles GmbH, Berlin, Germany) prior to the experiments. Slices of retinal tissues were transferred to the AFM and measurements were performed with a maximum force of 4–5 nN at an approach speed of 10 µm s⁻¹. Force-distance curves were taken every 15 µm in a raster scan across the whole apico-basal length of the retina (the average area covered was approximately 300  ×  200 µm²). In parallel, fluorescence images were acquired using an sCMOS camera (Zyla 4.2, Andor). Force-distance curves were fitted with the Hertz model [29] using a custom-written algorithm [2, 8, 10], and the apparent elastic modulus $K=E/\left(1-{{\nu}^{2}}\right)$ was extracted: $F=\frac{4}{3}\frac{E}{1-{{\nu}^{2}}}{{r}^{1/2}}{{\delta}^{3/2}}=\frac{4}{3}K{{r}^{1/2}}{{\delta}^{3/2}}$, with $F$  =  applied force, $E$  =  Young's modulus, $\nu$  =  Poisson's ratio, $r$  =  radius of the probe, and $\delta$  =  indentation depth. The algorithm was executed for various possible contact points in an iterative manner, and the best least-root-mean-square fit was chosen [8]. The curves were analyzed for the full indentation depth (i.e. for the maximum force applied). For further analysis, the retinal layers were segmented manually.

The confocal Brillouin microscope used in this study has been previously described [18]. The system uses a 532 nm laser with an optical power on the sample of 5 mW–10 mW and exposure time of 0.1 s–0.2 s. The laser light was focused by a 40  ×  objective (Olympus, NA  =  0.6) leading to an optical resolution of 1 µm (lateral) and 2 µm (axial). The scattered light was collected in epi-detection and sent to a two-stage VIPA-based Brillouin spectrometer featuring an EMCCD camera (Andor, IXon Du-897) [30–32]. To determine Brillouin frequency shifts, spectral data acquired by the camera were fitted by a Lorentzian function using a custom MATLAB code. 2D images were obtained by scanning the sample with a 3D translational stage (Prior Sci.). Using a reference arm, water and glass samples of known Brillouin frequency shifts were used for the long-term calibration of spectral data. A custom LabVIEW (National Instruments) code was used to control the motorized stage, shutters, and the CCD camera. Retinal layers were segmented manually.

Data will be provided upon request.

We first compared the morphological structure of retinae of two ruminant species, cow and sheep, using hematoxylin and eosin (H&E) stainings. In an H&E stain, nuclei appear dark (blue–purple), while cytoplasm and other structures, such as cellular processes, stain pink. Cross-sections of ovine (figure 1(a)) and bovine (figure 1(b)) retinae showed a very similar morphology.

Zoom In Zoom Out Reset image size

Photoreceptor cells accounted for the thickest layer, taking up almost half of the total retinal thickness. Densely packed outer (OS) and inner photoreceptor cell segments (IS; pink staining at apical side) projected from a thick layer of stacked photoreceptor cell somata (ONL), 5–7 nuclei in width. OS, IS, and ONL together form the photoreceptor cell layer (PRL). The underlying outer plexiform layer (OPL), containing mostly neuronal processes and their synapses, was the thinnest layer of the retina. The neighboring INL, containing somata of interneurons and Müller glial cells, was approximately 3–4 nuclei wide in both species (corresponding to ~one fifth of the total retinal thickness). The adjacent IPL was of similar width as the INL. Finally, the basal retinal ganglion cell layer (GCL) was a sparse, single-layered structure with individual ganglion cells (GCs) scattered across the NFL. Taken together, the retinae of sheep and cow are very similar in terms of nuclear density distributions and thus both good model systems to study ruminant retina mechanics.

To determine the mechanical properties of ovine retinae with cellular resolution, we performed AFM indentation measurements across retinal cross-sections in a raster scan with a step size of 15 µm (figure 2, supplementary movie 1). Fitting the Hertz model to force-distance curves, we determined the apparent elastic modulus K at each position. K is a measure of the tissue's local elastic stiffness.

Zoom In Zoom Out Reset image size

To correlate the AFM measurements with the underlying retinal structures, we treated retinae with MitoTracker Orange prior to AFM measurements. MitoTracker stains mitochondria in live cells and in the retina particularly Müller glial cells [28], allowing to visualize individual tissue layers (figure 2(a), supplementary movie 1). Plotting color-coded K values as a function of position in the measurement plane resulted in elasticity maps (figure 2(b)) [2, 8, 10, 17], revealing that ovine retinal cross-sections are mechanically highly heterogeneous, with alternating stiff and soft regions.

This finding was confirmed by comparing pooled data of each tissue layer (figure 3(a)). In all AFM experiments (n  =  3), we found significant differences between the mechanical properties of different retinal layers (P  <  0.001, Kruskal–Wallis ANOVA). The areas of highest stiffness corresponded to the PRL and the more basally located INL, i.e. the layers containing most cell bodies in the retina (figures 2 and 3(a), supplementary figures 2(a) and 3). Within the PRL, photoreceptor cell bodies as well as IS and OS were mechanically indistinguishable. With a median K of 330 Pa, the PRL was significantly stiffer than all other retinal layers (n  =  3; P  ⩽  4 ^* 10⁻⁴, Dunn's multiple comparison test). While OPL, IPL, and the NFL/GCL were mechanically similar, with median K values of 199 Pa, 153 Pa, and 157 Pa, respectively (P  ⩾  0.26), the IPL was significantly softer than the neighboring INL with K  =  216 Pa (P  =  3.9 ^* 10⁻²) (figure 3(a)).

Zoom In Zoom Out Reset image size

Having established that AFM can be used to distinguish different retinal layers by their mechanical fingerprints, we next compared the AFM measurements with a contactless optical method recently developed to measure tissue mechanics, Brillouin microscopy [18–20, 33] (supplementary figure 1(b)). Measurements of cross-sections of bovine retinae revealed a very similar elastic stiffness distribution in the tissue as found by AFM (figure 2). In both approaches, two comparatively stiffer regions were identified: the apical third of the retina, corresponding to the PRL, and a thin band located more basally, which co-localized with the INL. Plotting the median values of each pixel row in the heat maps shown in supplementary figure 3 (AFM) and figure 2(c) (Brillouin) resulted in line profiles of stiffness distributions within the ruminant retina (figures 2(d) and (e)). These plots confirmed an overall stiffness gradient in the tissue. The overall stiffness decreased from the apical to the basal retinal surface. Both plexiform layers showed a dip in elasticity, with lower K values and smaller Brillouin shifts, with an intermediate peak in stiffness coinciding with the INL.

To quantitatively compare the two techniques, let us consider the physics behind Brillouin and AFM interactions. In AFM, a cantilever with a spherical probe physically indents the sample. From the resulting force-distance curve, the apparent elastic modulus K can be calculated, which is related to the quasi-static Young's modulus E through the relationship $K=E/\left(1-{{\nu}^{2}}\right)$, where ν is the Poisson's ratio. In spontaneous Brillouin scattering, the light scattered from a phase-matched acoustic wave experiences a Doppler-like frequency shift due to the propagating soundwaves. In the experimental configuration illustrated in supplementary figure 1(b), the Brillouin frequency shift can be expressed as Ω $~=\left(2n/\lambda \right)\sqrt{{{M}^{'}}/\rho}$, where λ is the wavelength of light, n the refractive index of the tissue, M' the real part of the longitudinal elastic modulus, and ρ the density of the material. Therefore, Brillouin spectroscopy can provide direct information about the local longitudinal elastic modulus of a material (figure 4), i.e. the uniaxial stress–strain ratio at frequency Ω, using the equation $~{{M}^{\prime}}=\rho /4{{n}^{2}}{{\lambda}^{2}}{{ \Omega }^{2}}$. Assuming an average refractive index of the retina of about 1.358 [34] and a density of the cow retina of 1.033 g cm⁻³ [35], $\frac{\rho}{{{n}^{2}}}$ can be approximated by a constant ($=0.56\,\text{g}\,\text{c}{{\text{m}}^{-3}}$) within a tissue [22, 27].

Zoom In Zoom Out Reset image size

In crystalline materials, the relationship between the longitudinal modulus M derived from Brillouin measurements and Young's modulus E is straightforward, i.e. $M=E\left(1-\nu \right)/\left(1+\nu \right)\left(1-2\nu \right)$. On the other hand, in biological tissues, this relationship is not established. Biological materials are generally characterized by a low compressibility, i.e. their Poisson's ratio is close to 0.5; in the retina ν has been reported to be about 0.47 [36]. Furthermore, as all other biological tissues, neural tissue is nonlinearly elastic; its elastic modulus strongly depends on the frequency at which it is probed [5]. Thus, the longitudinal modulus M measured by Brillouin techniques at GHz frequencies is much higher than the quasi-static apparent elastic modulus, K, determined by AFM at frequencies in the ~Hz range.

Previously, we have empirically compared Brillouin measurements with different mechanical tests of biological tissues and biopolymers [18, 22, 27]. We found a strong correlation between longitudinal and Young's moduli, in a log–log linear relationship: log(M')  =  alog(E')  +  b, with a and b being material-dependent coefficients. This correlation suggests a power-law relationship between M' and E', consistent with the power-law frequency-scaling found in biological tissues and biopolymers up to kHz [37–39] and MHz [40]. Comparing AFM and Brillouin measurements of this study, we found a similar relationship between both moduli, allowing to convert the Brillouin shift into a quantitative assessment of elastic tissue stiffness (figure 4). The best fit was obtained for a  =  0.07 and b  =  9.22 (R²  =  0.99).

Recently, we developed an empirical mathematical model to estimate the local stiffness of murine spinal cord tissue based on histology data [10]:

$$\begin{eqnarray}&&{{K}_{c}}=a\times {{A}_{\text{norm}}}+b\times \Theta +c\end{eqnarray} \tag{ 1 }$$

where a, b and c are constants that depend on the material tested and on measurement parameters, A_norm is a measure of cell body density, and $\Theta$ a measure of axon orientation. In all retinal layers except the NFL, $\Theta$  =  0, as long axons are missing [16] and there is thus no dominant direction along which axons are aligned, simplifying this expression to

$$\begin{eqnarray}&&{{K}_{c}}=a\times {{A}_{\text{norm}}}+c\end{eqnarray} \tag{ 2 }$$

To test if this equation can be generally applied to CNS tissue, we determined local cell body densities in the largest retinal layers, ONL, INL, and IPL (figure 5(a)), and calibrated the model using the AFM results. As in the mouse spinal cord, tissue stiffness linearly scaled with the density of cell nuclei (R²  =  0.99; figure 5(b)), suggesting that mechanical heterogeneities in CNS tissue are largely governed by cellular structures, and that local CNS tissue stiffness can be estimated based on histology data.

Zoom In Zoom Out Reset image size