Single-nuclei RNA-seq on human retinal tissue provides improved transcriptome profiling

The retina is the innermost layer of tissue in the eyes of human and most other vertebrates. It receives the information of the visual images like the film of a camera, translates the images into neural signals, and transduces the signal to the brain. Three layers of neural cells (photoreceptor cells, bipolar cells, and ganglion cells) within the retina are comprised of seven major cell types; rod, cone, Müller glia cell, amacrine cell, horizontal cell, bipolar cell, and retinal ganglion cells, which create visual perception through functional cooperation. In all retinal diseases, it is the ultimate degeneration of the photoreceptors, the rods and cones, which results in blindness. Understanding all the individual cell types and how they contribute the basic neural circuitry in both the human and mouse retina will be key to elucidating mechanism of retinal degenerations. Here, we report to the best of our knowledge the first single-nuclei RNA-seq based transcriptomic study on human neural retinal tissue. We sequenced 6544 nuclei from six samples, and 4730 nuclei passed the quality filtering steps. The donor samples came from both the macular and peripheral regions, respectively, from three donors’ retina. None of the donors had any evidence of any ocular disease including retinal diseases as evidenced by medical record and post mortem imaging and phenotyping (Owen et al., IOVS in press). Unsupervised clustering was able to identify all seven cell populations. Known markers of neural retinal cell types were used to assign the clusters to these seven major retinal cell types, and thus, gene expression profiles for each cell type were obtained. With this dataset, the similarities and differences in cell proportion and gene expression between human and mouse were assessed. We highlighted the genes that demonstrated statistically significant differential distribution between human and mouse photoreceptors. In addition, the gene expression profiles of the same cell types between the macular and peripheral regions between human and mouse were compared. Specifically, as a means of validation the highly expressed genes in photoreceptor cells were found to be significantly enriched in known inherited retinal disease genes. Thus, we propose that this cell-specific gene list would serve as a prioritization during not only novel disease gene discovery, but cell-specific pathway gene analysis. This should facilitate a better understanding of retinal cell biology with the goal of cell-specific therapeutic interventions for retinal degenerations, as therapies are limited.


61
The transduction of visual signals occurs in a highly specialized tissue of the central  11 . It further encompasses a 1.5mm central region which 83 includes an avascular pit comprised of long-and medium-wavelength (L/M) cone 84 photoreceptors and a few short-wavelength (S) cones, but no rods 11 . This unique 85 distribution is essential both for high visual acuity and color vision perception in the retina. 86 The macula is responsible for central vision and is necessary for reading, recognition of 87 faces, and driving. The macula is surrounded by a region in the retina that is highly 88 enriched in rods constituting the peripheral retina that is functionally crucial for dim light 89 and peripheral vision 11 . These differences are not solely architectural, but also have 90 unique differentiation and downstream cellular connections indicating a distinct functional 91 and thus transcriptional profile of cells in the central vs the peripheral region [11][12][13] . The 92 differences between the macular and periphery has not yet been investigated at a single-93 cell transcriptomic level.   In this study we use single nuclei-RNA seq, here in referred to Nuc-seq which is an 111 improvement over the standard single cell-RNA seq especially for neuronal tissue [25][26][27][28] .     microscope with a robotic stage, to visualize wells containing single nuclei (see Table 1 183 for single cell capture number across different experimental repeats). Automated well 184 selection was performed using the CellSelect software (Wafergen Biosystems) which 185 identified nanowells containing single nuclei, and excluded wells with >1 nuclear, debris, 186 nuclei clumps or empty wells. The candidate wells were manually evaluated for debris or 187 clumps as an additional QC.  For mouse single-cell RNA-seq, the expression data was obtained from GSE63473 7 .

226
Matrices from seven P14 mice, GSM1626793-1626799, were used for analysis. Gene    Table 2). Gene ontology enrichment analysis of 274 biological process terms were performed with these DEGs (Supplement Table 3).   Table 2). These genes show significant 295 enrichment of biological process GO terms of visual perception, rhodopsin mediated 296 signaling pathway, maintenance of photoreceptor cells and regulation of rhodopsin 297 mediated signaling pathway (Figure 2A). This is largely consistent with the known function 298 of photoreceptor cells, which further validate our cluster assignment and DEG analysis.  Table 4).   Table 5). These genes were compared with the 333 122 and 299 human rod and cone DEGs and showed considerable overlap ( Figure 2D).

334
Of note, we found 19 genes as both mouse rod and cone DEGs, but only as human cone DEGs (Supplement Table 5), indicating that mouse models might not be that suitable for 336 this set of genes. Two of these genes, RPGRIP1 and RD3, appears as known human 337 IRD genes 42-51 and show high expression in human cone cells, compared with rod cells 338 ( Figure 2E). Patients with mutations in RPGRIP1 and RD3 were reported with LCA and 339 CRD, which showed a more severe cone phenotype than rod. In contrast, KO mouse 340 models of these two genes are reported to display RP phenotypes 51-53 , which were mostly 341 resulting from rod issues. Thus, we demonstrated that mouse models of specific retinal 342 degeneration associated genes do not faithfully recapitulate the human condition.   Table 6).  Table 7). As expected, robust 389 expression of most of these known IRD genes (197 of the 234) could be detected in our 390 dataset (Supplement Table 7). Additionally, the detected IRD genes are expressed at 391 significantly higher level than average ( Figure 4A, p-value = 2.27e-07, two-sample t-test).

392
Since the vast majority of known IRD associated genes affects photoreceptor cells  Given the significant enrichment of IRD associated genes in the cell type DEG set, it can 407 be potentially used to prioritize candidate IRD disease genes. To test this idea, the eye 408 phenotype for mice with photoreceptor cell DEG orthologous gene mutated is examined.

409
For the 103 photoreceptor DEGs that have not been associated with human retina 410 diseases, 49 have knock out mice recorded in the MGI database with non-lethal 411 phenotypes. Among them, 18 shown phenotypes in the visual system (Supplement Table   412 8), such as GNGT1, RDH8 and RCVRN. Protein encoded by GNGT1 is known to locate  Table 8). From those genes, we found a few of them  Table 8). These genes could potentially serve as 439 prioritization for further study of non-PR-related IRDs. The utility of single-nuclei RNA-seq on human retinal tissue is also assessed in our study. of this approach, although this limitation did not affect our study on the major cell types. 458 We found 299 highly expressed genes in human cones and looked for differences 459 between human rods and cones. Currently, cone cells are much less studied than rod 460 cells. This is because: first, the human cone cells are more different from mouse cone