Quantitative and Qualitative Evaluation of Photoreceptor Synapses in Developing, Degenerating and Regenerating Retinas

Abstract

Quantitative and qualitative evaluation of synapses is crucial to understand neural connectivity. This is particularly relevant now, in view of the recent advances in regenerative biology and medicine. There is an urgent need to evaluate synapses to access the extent and functionality of reconstructed neural network. Most of the currently used synapse evaluation methods provide only all-or-none assessments. However, very often synapses appear in a wide spectrum of transient states such as during synaptogenesis or neural degeneration. Robust evaluation of synapse quantity and quality is therefore highly sought after. In this paper we introduce QUANTOS, a new method that can evaluate the number, likelihood, and maturity of photoreceptor ribbon synapses based on graphical properties of immunohistochemistry images. QUANTOS is composed of ImageJ Fiji macros, and R scripts which are both open-source and free software. We used QUANTOS to evaluate synaptogenesis in developing and degenerating retinas, as well as de novo synaptogenesis of mouse iPSC-retinas after transplantation to a retinal degeneration mouse model. Our analysis shows that while mouse iPSC-retinas are largely incapable of forming synapses in vitro, they can form extensive synapses following transplantation. The de novo synapses detected after transplantation seem to be in an intermediate state between mature and immature compared to wildtype retina. Furthermore, using QUANTOS we tested whether environmental light can affect photoreceptor synaptogenesis. We found that the onset of synaptogenesis was earlier under cyclic light (LD) condition when compared to constant dark (DD), resulting in more synapses at earlier developmental stages. The effect of light was also supported by micro electroretinography showing larger responses under LD condition. The number of synapses was also increased after transplantation of mouse iPSC-retinas to rd1 mice under LD condition. Our new probabilistic assessment of synapses may prove to be a valuable tool to gain critical insights into neural-network reconstruction and help develop treatments for neurodegenerative disorders.

Keywords: circuit reconstruction; photoreceptor synapse; retinal degeneration; ribbon synapse; stem cell therapy; synapse quantification; synaptogenesis.

Figure 1

Overview of QUANTOS, a synapse evaluation method using a Naïve Bayes classifier. (A) Photoreceptor synapses are visualized by immune-staining of pre-synaptic marker: RIBEYE, and post-synaptic marker: mGluR6. Three to four replicate IHC Images from three P28 B6J mice each were used as training data for Ideal Synapse and Ideal Noise. The OPL area was manually isolated to train the Ideal Synapse, and the area outside the OPL was used to train the Ideal Noise. Scale bar = 4 μm. (B) IHC images were processed by custom made ImageJ Fiji macros. IHC images were segmented, and thresholded using the background intensity of each segment. The thresholded areas were then overlaid on the original IHC image to extract graphical parameters from ROIs. Details of image processing steps for DAPI, RIBEYE, and mGluR6 are described in Figures S1–S3. (C) Upper panel: The distribution of extracted parameters was estimated with either Kernel Density Estimation or Bounded Density Estimation to generate PDFs for Ideal Synapse and Ideal Noise. These PDFs were used to estimate likelihoods of each synaptic marker. Marker spatial density is used to calculate prior probability. Pre- and post-synaptic markers within 1.2 μm of each other (distance from center of mass) were considered as synapse candidates. Lower panel: Posterior probability of synapse candidates being either synapse or noise is estimated by multiplying prior probabilities and likelihoods of both pre- and post-synaptic markers. (D) Posterior probability of being synapses are estimated for each individual synapse candidates. Synapse candidates with more than 50% of posterior probability were classified as synapse. IHC, immunohistochemistry; IPL, inner plexiform layer; OPL, outer plexiform layer; ONL, outer nuclear layer; ROIs, regions of interest; PDFs, probability density functions.

Figure 2

Sensitivity and specificity of QUANTOS. (A) ROC curves of classifiers using different combinations of parameters on a P28 sample. ROC curves for each parameter are indicated with a magenta line. The ROC curve of distance and all parameters are shown in all panels for comparison. Dots indicate the results of manual counts by different observers (IHC image: n = 1). (B) Comparison of AUC between different combination of parameters on P28 sample. Whiskers indicate 95% confidence intervals. (C) ROC curves of classifiers using different combinations of parameters on P14 sample (IHC image: n = 1). (D) Comparison of AUC between different combination of parameters on P14 sample. Whiskers indicate 95% confidence intervals. (E) Example of an IHC image of B6J P28 mouse. Yellow box area is shown magnified in (E'). Scale bar = 10 μm. (E') upper panel: Yellow small dots indicate synapses detected by QUANTOS., middle panel: blue large dots indicate the Ground Truth (manually evaluated by an expert), lower panel: overlay image of both QUANTOS results and Ground Truth. Scale bar = 5 μm. (F) Pre- (left column) and post-synaptic marker (right column) coordinates detected by QUANTOS. Each row shows the synapse candidates, i.e., candidates with high synapse likelihood given different parameters. White dots represent all the markers detected in the Image Processing, and gray dots represents all the synapse candidates (pre- and post-synaptic markers within 1.2 μm), and colored dots represent the synapse candidates with higher likelihood of synapse than noise for different parameters. “all parameters” represents the combined likelihoods of all parameters and pre- and post-synaptic markers. “posterior probability” shows the marker pairs identified as synapses by QUANTOS, which are obtained from “all parameters” by taking into account the prior probability of synapse. ROC, receiver operation characteristics; AUC, area under the curve.

Figure 3

Qualitative and quantitative evaluation of developmental synaptogenesis by QUANTOS in mice reared in DD and LD conditions. (A) IHC images of B6J mice on different postnatal days. RIBEYE is the pre-synaptic marker expressed in photoreceptors and mGluR6 is the post-synaptic maker expressed in bipolar cells. Images in upper row show the overview morphology of OPL, and lower row show the magnification of OPL. Scale bar = 10 μm for upper row, 2.5 μm for lower row. (B) Result of synapse quantification of postnatal B6J reared under LD and DD conditions. Dots indicate the number of synapses detected in each IHC image. Shape of dots represents the mouse ID for each time point. Dots of each time point were horizontally jittered for better visualization. The dark color-filled area shows the estimated range of mean number of synapses, and the pale color-filled area represents the estimated range of synapse numbers from each IHC image. (n = 3 for P7, P10, P35, and n = 4 for P14, 21, 28 samples. 3–4 replicates were taken from each mouse as indicated by the shape of markers). (C) Posterior distributions of modified Gompertz model parameters with 89% confidence interval. (D) Difference of posterior distributions of parameters between LD and DD conditions. (E) Developmental synaptogenesis was parameterized with the modified Gompertz's growth curve which has three parameters; the maximum number of synapses (A), maximum rate of synaptogenesis (μ_M), and the onset of synaptogenesis (λ). (F) 2D histograms of all synapse candidates on different postnatal days, with log synapse likelihood on the x axis, and log noise likelihood on the y axis. Synapse candidates on the left-upper side are more likely to be noise, and the ones on the right-lower side are more likely to be synapses. (G) All synapses detected by QUANTOS were averaged to visualize the characteristics of synapses on different postnatal days and different rearing conditions. (H) Radial profile plots of averaged synapses. The plots show the signal intensity in relation to the center coordinates of pre- and post-synaptic markers. Colors indicate different postnatal days. IHC, immunohistochemistry; INL, inner nuclear layer; OPL, outer plexiform layer, ONL; outer nuclear layer; LD, cyclic light; DD, constant dark; P, postnatal day; HDI, high density interval.

Figure 4

Rearing light conditions alter developmental synaptic function. (A) An example of mERG recording. Retinas flat-mounted on the 60-channel probe were stimulated with a mesopic light pulse. The red box is a magnified view of a single channel recording trace, showing a typical waveform with an a-wave and a b-wave. (B) Upper panels show histograms of b-wave amplitudes of wildtype P14 mice reared under different light conditions (n = 5 for DD and n = 4 for LD and LL) and the posterior predictive check of the statistical model used to analyze the data is shown in the lower panels. (C) Posterior distributions of mean b-wave amplitude. (D) Estimated impact of light on mean b-wave amplitudes. (E) Posterior distributions b-wave amplitude SD. (F) Estimated impact of light on SD of mean b-wave amplitudes. mERG, micro electroretinography; HDI, high density interval; SD, standard deviation.

Figure 5

QUANTOS evaluation of synapses during photoreceptor degeneration. (A) IHC images of rd1 mice retinas on different postnatal days. Images in upper row show the overview morphology of OPL, and lower row show the magnification of OPL. Scale bar = 10 μm for upper row, 2.5 μm for lower row. (B) The number of synapses detected on different postnatal days in rd1 mice, accompanied by B6J LD data for comparison (n = 3 for each postnatal day of rd1. 3–4 replicates were taken from each mouse as indicated by the shape of markers). (C) Number of photoreceptor cells estimated form IHC images on different postnatal days of rd1 mice retinas, accompanied by B6J data for comparison. (D) 2D histograms of all synapse candidates on different postnatal days, with log synapse likelihood on the x axis, and log noise likelihood on the y axis. (E) Averaged images of all synapses detected by QUANTOS show the characteristics of synapses on different postnatal days and different rearing conditions. Scale bar = 0.5 μm. (F) Radial profile plot of averaged synapses. This plot shows the intensity of signals in relation to the center coordinates of pre- and post-synaptic markers. Colors indicate different postnatal days. IHC, immunohistochemistry; INL, inner nuclear layer; OPL, outer plexiform layer, ONL; outer nuclear layer; SD, standard deviation.

Figure 6

QUANTOS detects de novo synapses after miPSC-retina transplantation and shows that light enhances synaptogenesis. (A–C) Example IHC images of rd1 mice after miPSC-retina transplantation on PT 14, 30, 60. Bottom panels show magnified images of some synapse candidates. Scale bar = 10 μm (D) Number of synapses of rd1 mice before and after transplantation of miPSC-retina under different rearing light conditions. (5 and 4 retinal organoids were sampled for in vitro dd25 and dd36, respectively. n = 4 for PT10 LD, n = 3 for PT14 LD, n = 2 for PT14 DD, n = 5 for PT30 LD, n = 4 for PT30 DD, n = 5 for PT60 LD, n = 4 for PT60 DD. 3–4 replicates were taken from each mouse as indicated by the shape of markers). (E) Estimated mean number of synapses per photoreceptor on PT 14, 30, and 60. (F) Difference of estimated mean number of synapses per photoreceptor between DD and LD. (G) 2D histograms of all synapse candidates on different postnatal days, with log synapse likelihood on the x axis, and log noise likelihood on the y axis. (H) Average synapse of rd1 mice before and after miPSC-retina transplantation. All synapses detected by QUANTOS were averaged from different time points, respectively. Scale bar = 0.5 μm. (I) Radial profile plot of averaged synapses. This plot shows the intensity of signals in relation to the center coordinates of pre- and post-synaptic markers. Colors indicate different postnatal days. Data of B6J is presented together for comparison. IHC, immunohistochemistry; PT, post-transplantation day; LD, cyclic light; DD, constant dark; dd, differentiation day; INL, inner nuclear layer; OPL, outer plexiform layer, ONL; outer nuclear layer.

Figure 7

QUANTOS can compare the relative maturation of synapses formed during development, degeneration, and regeneration of the retina. (A–C) 2D histograms showing the log likelihood of mature synapse on the x axis, and the log likelihood of immature synapse on the y axis. (A) Synapse maturation of B6J mice reared under LD or DD conditions with representative IHC images of synapses from P10 DD and P28 LD. (B) Synapse maturation of rd1 mice with representative IHC images of mature and immature synapses are presented as examples. (C) Synapse maturation of rd1 mice after miPSC-retina transplantation with representative IHC images of mature and immature synapses. (D) Synapses of rd1 mice after miPSC-retina transplantation with pre- and post-synaptic maker log mature/immature likelihoods displayed separately. IHC, immunohistochemistry; LD, cyclic light; DD, constant dark.