Figure 3. Modulation of the principal cell transcriptome across the male reproductive tract.

(A) Clusters from the overall dataset (Figure 2A), with the principal cell clusters extracted for reclustering highlighted in black. (B) Reclustering of extracted principal cells, visualized by t-SNE. Distinct colors highlight the 15 resulting principal cell clusters. See also Figure 3—figure supplements 1–4. (C) Reclustered principal cells, as in panel (B), with cells colored according to anatomical origin. (D) Heatmap showing the top 20 genes enriched in each of the principal cell clusters. Based on highly enriched marker genes, our epididymis clusters from 1 to 11 (ignoring the four vas deferens clusters) correspond to the following segments from Johnston et al., 2005: segment 1 (initial segment), segments 1–2, segments 3–4, segment 5 (early), late segment 5/segment 6, segment 7, late segment 7, segments 8–9, segment 9, late segment 9 (some segment 10 markers), segment 10. (E) Expression of key marker genes across all principal cells, visualized as in Figure 2C.

Figure 3—figure supplement 1. Gene expression heterogeneity within principal cell subclusters.

In order to explore cell-to-cell variability in mRNA abundance between individual cells comprising any given principal cell subcluster, we set out to identify the fraction of cells in each subcluster expressing any given gene. However, the low efficiency of mRNA capture in scRNA-Seq makes it difficult to distinguish ‘true zeroes’ – cases where a given cell does not express a gene – from technical dropouts resulting from failure to capture any molecules of the gene in question. To address this, we used DESingle (Miao et al., 2018) to estimate the true fraction of expressing cells for any given gene by inferring statistical likelihood of technical dropouts. Full dataset for this analysis is included as Supplementary file 3. (A) Principal cell clusters 1-15, reproduced from Figure 3B. (B) Using the output of DESingle, three plots show the relationship between gene expression level and fraction of expressing cells, using the principal cell subcluster corresponding to the initial segment (C1) for illustration. Left panel shows average expression level (across all cells in the cluster) on the x axis, expressed as UMIs normalized to parts per million, compared to expression level only in the subset of cells that detectably express the gene. At high expression levels, these values converge, indicating that extremely highly expressed genes are generally consistently detected throughout the cell population, whereas at lower average expression many genes fall above the diagonal, highlighting genes expressed in only a subset of cells. Middle and right panels compare a gene’s expression level (either confined only to expressing cells, or across all cells in the cluster, as indicated) to the fraction of expressing cells. Clu is shown in all three panels to highlight an example of high cell-to-cell variation for a gene known to exhibit patchy expression in the caput epididymis (Hermo et al., 1991). (C) Scatterplots for all 15 principal cell subclusters, comparing gene expression level only in expressing cells (x axis) to fraction of expressing cells (y axis). Substantial cell-to-cell heterogeneity can be observed across all subclusters – in general, genes located in the lower right area of these graphs represent highly expressed genes with unusually high cell-to-cell variability. See also Supplementary file 3. (D-E) Illustrative examples of genes exhibiting similar abundance in expressing cells (plotted in left panel of (D) for the first five principal cell clusters), but with different behaviors across cells (right panel, (D)). Expression across all principal cell subclusters is illustrated in (E) for all four genes. Wfdc10 and Crisp1 are both consistently expressed in the majority of cells in all five subclusters, and are typical examples of highly expressed genes. Rnase9 is a marker of clusters 4–5 and is penetrantly-expressed in these clusters. Importantly, a small number of Rnase9-positive cells can be detected in the more proximal clusters 1–3, and, interestingly, these rare cells express Rnase9 at similar levels to cells in clusters 4–5. Finally, Cst12 is one of the markers of cluster 2, yet even in this cluster only a small subpopulation of cells express this gene.

Figure 3—figure supplement 2. Region-specific expression of multi-gene family members.

(A) Heatmap showing expression of individual members of several multi-gene families (Rnase, Lcn, Cst, Defb, Wfdc, Spint, Spink, Serpin) across the 15 principal cell subclusters from Figure 3. (B) Three examples of relevant multigene clusters exhibiting region-specific expression of individual cluster members.

Figure 3—figure supplement 3. High levels of oxidative energy production in principal, clear, and muscle cells.

Heatmap shows expression of various nuclear-encoded mitochondrial genes (Mrpl, Mrps, Ndufa, Cox, Uqcr gene families) across the entire 21 cluster dataset from Figure 2A. Clusters expressing high levels of nuclear-encoded mitochondrial genes included all principal cell clusters, both muscle clusters, and the clear cells (leftmost 12 columns), while low expression of these genes was observed in basal, immune, and stromal cell types (rightmost nine columns).

Figure 3—figure supplement 4. Apical localization of Abcg2 in the vas deferens.

Left panels: immunostaining for Adm (a basal cell marker, in red) and Abcg2 (specific to vas deferens principal cells, in green), and counterstained with DAPI, in the epididymis. Images from the epididymis show Adm-positive basal cells, with no Abcg2-positive cells, as expected. Along with secondary only controls (not shown), the absence of Abcg2 staining in the epididymis provides additional negative controls for the other immunostaining studies in this manuscript. Right panels: vas deferens images show Abcg2 and DAPI staining, confirming widespread principal cell expression and predicted apical staining of Abcg2.