Evolution of neuronal cell classes and types in the vertebrate retina

The basic plan of the retina is conserved across vertebrates, yet species differ profoundly in their visual needs1. Retinal cell types may have evolved to accommodate these varied needs, but this has not been systematically studied. Here we generated and integrated single-cell transcriptomic atlases of the retina from 17 species: humans, two non-human primates, four rodents, three ungulates, opossum, ferret, tree shrew, a bird, a reptile, a teleost fish and a lamprey. We found high molecular conservation of the six retinal cell classes (photoreceptors, horizontal cells, bipolar cells, amacrine cells, retinal ganglion cells (RGCs) and Müller glia), with transcriptomic variation across species related to evolutionary distance. Major subclasses were also conserved, whereas variation among cell types within classes or subclasses was more pronounced. However, an integrative analysis revealed that numerous cell types are shared across species, based on conserved gene expression programmes that are likely to trace back to an early ancestral vertebrate. The degree of variation among cell types increased from the outer retina (photoreceptors) to the inner retina (RGCs), suggesting that evolution acts preferentially to shape the retinal output. Finally, we identified rodent orthologues of midget RGCs, which comprise more than 80% of RGCs in the human retina, subserve high-acuity vision, and were previously believed to be restricted to primates2. By contrast, the mouse orthologues have large receptive fields and comprise around 2% of mouse RGCs. Projections of both primate and mouse orthologous types are overrepresented in the thalamus, which supplies the primary visual cortex. We suggest that midget RGCs are not primate innovations, but are descendants of evolutionarily ancient types that decreased in size and increased in number as primates evolved, thereby facilitating high visual acuity and increased cortical processing of visual information.


Supplementary Note 1
Approximate morphologies and stratification of four types of cone bipolar cells and four types of RGCs in mouse and primate retinas were drawn based on the illustrations from previous literature listed in the table below.For a detailed description of the FLDA method, please refer to our previous work [5].In brief, FLDA is a method that projects high-dimensional gene expression data from cells with multiple categorical attributes into a low-dimensional space where each axis captures the variation along one attribute while minimizing co-variation with other attributes.
In this study, we used FLDA to analyze three categorical attributes of retinal neurons: response polarity (ON vs. OFF), response kinetics (transient vs. sustained), and species (mouse vs. primate).Let's use A, B, and C to represent these attributes.i, j, k denote the indices of attributes A, B, and C, and a, b, c are the number of categories in attributes A, B, and C. n ijk is the number of cells in the category combination ijk.x ijkl is the gene expression vector of the lth cell in the category combination ijk.
The covariance matrix of total variance can be decomposed as: where and Σ T is the total covariance matrix, and Σ A , Σ B , and Σ C are covariance explained by attributes A, B and C respectively.Σ e is the residual variance that is not explained by these attributes. Here, and Our objective was to find a projection that maximizes the variance of attribute A while minimizing the variances of attributes B and C. Specifically, we aimed to find u * that maximizes the following equation: where and This optimization problem is commonly referred to as a generalized eigenvalue problem [2].Here, N A is symmetric but not necessarily positive definite, and N e is positive definite.When N e is invertible, the eigenvector u * associated with the largest eigenvalue of N −1 e N A is selected.In this study, we identify the eigenvector with the largest eigenvalue of N −1 e N A , which we refer to as the FLDA eigenvalue for the attribute A. This FLDA eigenvalue measures how much variance of the corresponding attribute (A) is captured compared to the variances of other attributes (B, C).The eigenvector u * can be normalized to have a unit length.The elements within the unit vector represent the relative weights of the corresponding genes.
Similarly, to find a low-dimensional representation aligned with a categorical attribute B, we maximized the objective: where and to find a low-dimensional representation aligned with a categorical attribute C, we maximized the objective: where
Data was preprocessed and normalized as described in refs.[6] and [4].Briefly, transcript counts within each column of the count matrix (genes × cells) were normalized to sum to the median number of transcripts per cell, resulting in normalized counts Transcripts-per-million (T P M ij ) for gene i in cell j.We used a log-transformed expression matrix E ij = ln(T P M ij + 1) for further analysis.
We next identified 7779 high-variance genes (HVGs) using an approach that fits a relationship between the mean and coefficient of variation of gene expression [1,3].Next, we performed principal component analysis (PCA) on the dataset to remove multicollinearity.Finally, we analyzed the resulting PCs x cells matrix using FLDA [5].
In order to determine the mouse RGC types that best match the four predominant primate RGC types: ON/OFF midgets and ON/OFF parasols, we selected 20 candidates of mouse types with known polarity and kinetics based on previous studies (Supplementary Table 4).We drew all possible combinations of four RGC types from this set (n=432), and for each combination, we performed FLDA and calculated the eigenvalue corresponding to the polarity and the kinetics axes.We ranked these combinations based on their FLDA eigenvalues and identified the combination with the highest eigenvalue as the best match (Fig. 5).
14) N is the total number of cells, and a − 1, b − 1, c − 1, and N − a − b − c + 2 are the degrees of freedom of the corresponding terms.

Table 1 :
A list of figures in previous reports that were used to draw rough morphology of bipolar cells and RGCs depicted in Fig.5b.