Different lineage contexts direct common pro-neural factors to specify distinct retinal cell subtypes

Abstract

How astounding neuronal diversity arises from variable cell lineages in vertebrates remains mostly elusive. By in vivo lineage tracing of ∼1,000 single zebrafish retinal progenitors, we identified a repertoire of subtype-specific stereotyped neurogenic lineages. Remarkably, within these stereotyped lineages, GABAergic amacrine cells were born with photoreceptor cells, whereas glycinergic amacrine cells were born with OFF bipolar cells. More interestingly, post-mitotic differentiation blockage of GABAergic and glycinergic amacrine cells resulted in their respecification into photoreceptor and bipolar cells, respectively, suggesting lineage constraint in cell subtype specification. Using single-cell RNA-seq and ATAC-seq analyses, we further identified lineage-specific progenitors, each defined by specific transcription factors that exhibited characteristic chromatin accessibility dynamics. Finally, single pro-neural factors could specify different neuron types/subtypes in a lineage-dependent manner. Our findings reveal the importance of lineage context in defining neuronal subtypes and provide a demonstration of in vivo lineage-dependent induction of unique retinal neuron subtypes for treatment purposes.

Graphical abstract

Figure 1.

Major neurogenic lineages in the zebrafish retina. (A) Schematic of zebrafish retina structure. OPL, outer plexiform layer. IPL, inner plexiform layer. (B) Schematic of the mMAZe construct. (C) The working flow of lineage analysis of 48-hpf RPCs using mMAZe. (D) Schematic of the atoh7:Switch plasmid (pAtoh7:Switch). (E) The working flow of lineage analysis of atoh7⁺ PRCs using atoh7:Switch. (F) Representatives of major neurogenic lineages traced by mMAZe. (G) Representatives of major neurogenic lineages traced by atoh7:Switch. (H–L) The top four neurogenic lineages, which were analyzed using mMAZe and produce RGCs (H), ACs (I), BCs (J), PRs (K), or HCs (L) are listed. Major neurogenic lineages of each neuron type (frequency >15%) are highlighted in blue. (M–O) The top four lineages, which were analyzed using atoh7:Switch and produce RGCs (M), ACs (N), or BCs (O) are listed. Major neurogenic lineages of each neuron type (frequency >15%) are highlighted in red. (P) Summary graph of six major neurogenic lineages. Scale bars, 10 µm.

Figure S1.

Characteristics of lineages traced by mMAZe and atoh7:Switch. (A) Cell type distribution of lineages traced by mMAZe (1,595 cells in total) and atoh7:Switch (1,302 cells in total). (B) Size (the number of cells per lineage) distribution of lineages traced by mMAZe (n = 511) and atoh7:Switch (n = 484). (C) Top4 lineages traced by atoh7:Switch which produce PR(s). (D) Top4 lineages traced by atoh7:Switch which produce HCs. (E–G) Atoh7 expression in AC-BC lineages. (E) Schematics of MAZe-mCherry. (F) Workflow to analyze atoh7 expression in AC-BC. Sparse mCherry expression was achieved by heat shock at 48 hpf. Spatially isolated AC-BC lineages were analyzed at 72 hpf. (G) Atoh7 expression (labeled by GFP) in AC-BC lineages. Image in yellow rectangle was zoomed in on the right panel. All images presented are in Z-stack. Scale bar, 10 µm.

Figure S2.

AC and BC subtypes. (A) Signal of Tg(gad1b:EGFP) and Tg(glyt1:EGFP). (B) Verify Tg(gad1b:EGFP) by double immunostaining of EGFP and gad65/67. 142/144 EGFP⁺ ACs were gad65/67⁺. Lower panels show the zoom-in result, and the colabeled cells are indicated by asterisks. (C) Verify Tg(glyt1:EGFP) by RNA in situ hybridization of glyt1 with immunostaining of EGFP. 130/132 EGFP⁺ ACs were glyt1⁺. Lower panels show the zoom-in result, and the colabeled cells are indicated by asterisks. (D and E) AC morphology in AC-2PR and AC-BC lineages, which were traced by mMAZe (D) and distribution summary is shown in E. (F) An example illustrates sequential photoconversion. BC 1 was determined as ON-subtype after the first photoconversion, and then BC 2 was determined as ON-subtype after the second photoconversion. (G and H) Atoh7⁺ BC subtypes were determined by photoconverting kaede-green to kaede-red (G). The distribution summary is shown in H. (I and J) All BCs from BC-MC lineages were ON- or ONOFF-subtype (I), and the statistical results are shown in J. All images are presented in Z-stacks. Cell terminals are indicated by arrowheads. Scale bar, 10 µm.

Figure 2.

AC and BC subtypes in major neurogenic lineages. (A and B) Analysis of GABAergic ACs in AC-2PR and AC-BC lineages. Representative images are shown in A. Merged channels and separated channels of ACs in yellow rectangles are shown in the right panels. The statistical results are summarized in B. (C and D) Analysis of glycinergic ACs in AC-2PR and AC-BC lineages. Representative images are shown in C. Merged channels and separated channels of ACs in yellow rectangles are shown in the right panels. The statistic results are summarized in D. (E–G) Analysis of BC subtypes in AC-BC and BC-BC lineages traced by mMAZe. Representative images of BC subtypes in E, statistical results of BC subtypes in AC-BC (F), and BC-BC (G). (H) Summary graphic of AC and BC subtypes in different lineages. Scale bar, 10 µm.

Figure 3.

AC respecification in AC-2PR and AC-BC lineages after ptf1α knockout (KO). (A) Signal of Tg(ptf1α:GFP) in control and after knocking out ptf1α. (B and C) Verification of ptf1α knockout efficiency through injecting four CRISPR/Cas9 ribonucleoprotein complexes in G0 zebrafish. (B) Examples of eight alleles around targeted sites of sgRNA 1 and sgRNA 4. The first line is the WT sequence, sgRNA targeted sites are highlighted in blue, and mismatches are highlighted in red. (C) Summary of characteristics of 59 ptf1α alleles analyzed. (D) The workflow to analyze AC respecification. (E) 9/10 ACs in AC-2PR lineages were respecified as two PRs, and the resulting lineages were 2PR^res-2PR^norm (left panel). 15/17 ACs in AC-BC lineages were respecified as single BC, and the resulting lineages were BC^res-BC^norm (right panel). Zoom-in images within yellow rectangles are shown in the right panels. (F) The summary graphic illustrates the AC respecification in AC-2PR and AC-BC lineages after ptf1α knockout. res, respecified; norm, normal. Scale bar, 10 µm.

Figure 4.

Defining RPC heterogeneity by scRNA-seq. (A) The workflow of scRNA-seq of 48-hpf retina and primary data analysis. (B) Four clusters of undefined RPCs in G2/M phase in A. (C) Top feature TFs of Clusters A–D in B. The size of each circle is the percentage of cells expressing the marker in each cluster, and its intensity represents the scaled expression level. (D) A schematic showing the experimental design to enrich atoh7⁺ RPCs. (E) Gene expression pattern of atoh7⁺ RPCs. Genes listed are featured TFs of 48-hpf RPCs in C. Clusters 1–3 show similar expression patterns as Clusters A–C of 48-hpf RPCs and are labeled as A’, B’, and C’.

Figure S3.

Single-cell RNA sequencing of 48-hpf retina. (A) The t-SNE plot of the 3,587 qualified cells of zebrafish retina at 48 hpf and the annotation. (B) Featured genes of each cell cluster in A. (C) Cell-cycle distribution of cells in A. Cells during the same cell-cycle phase tend to cluster together. (D) GO analysis of top featured genes of 48-hpf RPCs showed the enrichment of transcription factors. HLH, helix-loop-helix. (E) Gene expression pattern of the three populations of 48-hpf early RPCs. Genes listed are the featured TFs of 48-hpf RPCs in (Fig. 4 C).

Figure S4.

Single-cell RNA sequencing of atoh7⁺ RPCs. (A) Time-lapse result of Tg(atoh7:turboGFP-dest1::atoh7:gapRFP) retina. TurboGFP-dest1 signal appears and disappears earlier than that of gapRFP. (B) FACS analysis of retinal cells from the Tg(atoh7:turboGFP-dest1::atoh7:gapRFP) embryos. The GFP⁺RFP⁻ cells for scRNA-seq are highlighted. (C) The t-SNE plot showing the clustering result of GFP⁺RFP⁻ cells in B and the annotation. (D) Featured genes of each cell cluster in C. (E) Cell-cycle distribution of cells in C. Cells during the same cell cycle phase tend to cluster together. (F) Four clusters of atoh7⁺ G2/M RPCs. (G) Expression patterns of cluster 2–specific (onecut1, pou2f2a, myca) and cluster 3–specific (vsx1, olig2, pou3f1) TFs of cells in F.

Figure 5.

Lineage tracing of vsx1⁺ and OC1⁺ RPCs. (A and B) Schematics of vsx1:Gal4 (A) and OC1:Gal4 (B) BAC constructs. (C) Vsx1⁺ cells and OC1⁺ cells in the retina at 3 dpf. (D and E) Lineage tracing of vsx1⁺ RPCs by the photoconversion of kaede-green to kaede-red at 48–54 hpf. Representative images are shown in D. The statistical result is shown in E. (F and G) Representative images (F) and the statistical result (G) show vsx1⁺ AC subtypes. (H and I) Lineage tracing of OC1⁺ RPCs by collecting spatially isolated lineages. Representative images are shown in H. The statistical result is shown in I. (J and K) Representative images (J) and the statistical result (K) show OC1⁺ AC subtypes. (L) A summary graphic illustrates lineage bias of vsx1⁺ and OC1⁺ PRCs. Scale bar, 10 µm.

Figure 6.

Different chromatin accessibility dynamics of vsx1 and OC1 in 48-hpf RPCs. (A) A t-SNE plot of 48-hpf RPCs based on their chromatin accessibility. Clusters are annotated by their specific chromatin accessibility. (B) A dot plot of gene activity of the clusters in A. (C) A t-SNE plot of 48-hpf RPCs based on their chromatin accessibility. Cells are colored by their pseudo-time. (D) A schematic illustrates the developmental trajectory and lineage markers of 48-hpf RPCs. (E and F) Coverage plots of proximal (highlighted in red) and distal (highlighted in blue) elements of vsx1 (E) and OC1 (F) in different clusters of 48-hpf RPCs. (G) A summary schematic shows the different chromatin accessibility dynamics of vsx1 and OC1 in 48-hpf RPCs.

Figure S5.

ScATAC-seq of 48-hpf retina. (A and B) The clustering result of 48-hpf retinal cells. Clusters are shown in a t-SNE plot (A) and annotated by their specific gene activities. Gene activity patterns of all clusters are shown in a dot plot (B). (C and D) Clustering results of all RPCs (C) and atoh7^open RPCs (D). (E) Correlation among the five RPC populations of 48-hpf scATAC-seq data (Fig. 6 A) and Clusters A–D of 48-hpf scRNA-seq data (Fig. 4 B) with the minimal prediction scores of 0.3. (F and G) Cicero co-accessibility among elements surrounding vsx1 (F, top) and OC1 (G, top). Peak positions are marked by black bars. The proximal opening (highlighted in red) and distal opening (highlighted in blue) in the 48-hpf RPCs are shown in the coverage plots. The proximal element of OC1 (marked by the dashed box) is magnified (G, right).

Figure 7.

Overexpression of atoh7, otx2, and ptf1α in vsx1⁺ and OC1⁺ RPCs. (A) Representative images of vsx1⁺ cells (labeled by tdTomato-NLS) after overexpressing atoh7, otx2, and ptf1α in vsx1⁺ RPCs with no TF overexpression as the blank. Tg(ptf1α:GFP) was used to indicate cell body location. (B) Cell type distribution of vsx1⁺ cells in A. The significant increase of each neuron type compared with blank is shown. (C and D) Vsx1⁺ AC subtypes after ptf1α overexpression. Representative image of glyT1⁺ ACs is shown in C. Zoom-in images in the yellow rectangle are shown on the right side. Statistic distribution of vsx1⁺ AC subtypes is shown in D. (E) Representative images of OC1⁺ cells (labeled by tdTomato-NLS) after overexpressing atoh7, otx2, and ptf1α in OC1⁺ RPCs with no TF overexpression in the blank. (F) Cell type distribution of OC1⁺ cells in E. The significant increase of each neuron type compared with blank is shown. (G and H) OC1⁺ AC subtypes after ptf1α overexpression. The image of a gad1b⁺ AC is shown in G. Zoom-in images in the yellow rectangle are shown on the right side. Statistical distribution of OC1⁺ AC subtypes is shown in H. (I) A summary graph. Scale bar, 10 µm. **, P < 0.001; ***, P < 0.0001 (Fisher's exact test).