A lineage-resolved molecular atlas of C. elegans embryogenesis at single cell resolution

Abstract

C. elegans is an animal with few cells, but a striking diversity of cell types. Here, we characterize the molecular basis for their specification by profiling the transcriptomes of 86,024 single embryonic cells. We identify 502 terminal and pre-terminal cell types, mapping most single cell transcriptomes to their exact position in C. elegans’ invariant lineage. Using these annotations, we find that: 1) the correlation between a cell’s lineage and its transcriptome increases from mid to late gastrulation, then falls dramatically as cells in the nervous system and pharynx adopt their terminal fates; 2) multilineage priming contributes to the differentiation of sister cells at dozens of lineage branches; and 3) most distinct lineages that produce the same anatomical cell type converge to a homogenous transcriptomic state.

One sentence summary:

With the transcriptomes of 86,024 single cells from C. elegans embryos ranging from gastrulation to the start of cuticle synthesis, we reconstruct the cell lineage and characterize the differentiation of most terminal cell types.

To understand how cell fates are specified during development, it is essential to know the temporal sequence of gene expression in cells during their trajectories from early uncommitted precursors to differentiated terminal cell types. Gene expression patterns near branch points in these developmental trajectories can help identify candidate regulators of cell fate decisions (1). Single cell RNA sequencing (sc-RNA-seq) has made it possible to obtain comprehensive measurements of gene expression in whole animals (2–7) and embryos (8–14). sc-RNA-seq profiling of multiple developmental stages in a time series can be particularly informative, as algorithms can use the data to reconstruct the developmental trajectories followed by specific cell types. However, confounding factors can generate misleading trajectories. For example, progenitor cell populations with distinct lineage origins may be conflated if their transcriptomes are too similar, and abrupt changes in gene expression can result in discontinuous trajectories. Thus, information from independent assays is necessary to conclusively validate an inferred trajectory as an accurate model of development.

Here, we comprehensively reconstruct and validate developmental trajectories for the embryo of the nematode worm Caenorhabditis elegans. C. elegans develops through a known and invariant cell lineage from the fertilized egg to an adult hermaphrodite with 959 somatic cells (15, 16), which creates the potential for a truly comprehensive understanding of its development. Using sc-RNA-seq, the known C. elegans lineage, and imaging of fluorescent reporter genes (17, 18), we produce a lineage-resolved single cell atlas of embryonic development that includes trajectories for most individual cells in the organism. Our atlas expands on previous studies of the earliest embryonic blastomeres (19, 20), covering 87% of embryonic lineage branches.

We use this dataset to quantitatively model the relationship between the cell lineage and the temporal dynamics of gene expression. We find that during gastrulation, lineage distance between cells is a strong predictor of transcriptome dissimilarity. The strength of this correlation increases from the middle to the end of gastrulation. After gastrulation, expression patterns of closely related cells diverge as they adopt their terminal cell fates. Body wall muscle, hypodermis, and the intestine are exceptions to this trend, as they are produced by semi-clonal lineage clades that maintain within-clade transcriptomic similarity. In the ectoderm, the final two rounds of cell division produce distinct neuron and glia cell types, which rapidly differentiate, often resulting in discontinuities in computational reconstructions of their developmental trajectories. In several cases, the transcriptomes of distant lineages converge as they adopt the same terminal cell fate, and at the same time diverge from their close relatives in the lineage.

Our ability to reconstruct these complex gene expression dynamics highlights both the utility of the known C. elegans lineage and the challenges that will be faced when trying to use single cell RNA sequencing to reconstruct the lineages of other organisms.

Single-cell RNA-seq of C. elegans embryos

We sequenced the transcriptomes of single cells from C. elegans embryos with the 10x Genomics platform. We assayed loosely synchronized embryos enriched for pre-terminal cells as well as embryos that had been allowed to develop for ~300, ~400, and ~500 minutes after the first cleavage of the fertilized egg. We processed the datasets with the Monocle software package (21). After quality control, the final integrated dataset contained 86,024 single cells, representing a more than 60x oversampling of the 1,341 branches in the C. elegans embryonic lineage.

We estimated the embryo stage of each cell by comparing its expression profile with a high-resolution whole-embryo RNA-seq time series (22) (Fig. S1). We then visualized the data with the Uniform Manifold Approximation and Projection (UMAP) (23, 24) algorithm, which projects the data into a low-dimensional space and is well suited for data with complex branching structures (24). We found that trajectories (25) in the UMAP projection reflect a smooth progression of embryo time (Fig. 1A), with cells collected from later time points usually occupying more peripheral positions (Fig. 1B). Unique transcripts per cell, as estimated with Unique Molecular Identifiers (UMIs), decreased with increasing embryo time throughout the period of embryonic cell division, consistent with decreasing physical cell size (Fig. S2). These observations suggest that UMAP trajectories corresponded to developmental progression and that embryo time estimates are a reasonable proxy for developmental stage for most cells. Approximately 75% of the cells recovered (64,384 cells) were from embryos spanning 210–510 minutes post first cleavage, corresponding to mid-gastrulation (~190 cell stage) to terminal differentiation (3-fold stage of development) (Fig. 1C); however, cells were also recovered from earlier embryos (< 210 minutes, 9,886 cells), and later embryos (> 510 minutes, 11,754 cells).

Fig 1. UMAP projection shows tissues and developmental trajectories in C. elegans embryogenesis.

(A) UMAP projection of the 81,286 cells from our sc-RNA-seq dataset that passed our initial QC. This UMAP does not include 4,738 additional cells that were initially filtered, but were later whitelisted and included in downstream analyses. Color indicates the age of the embryo that a cell came from, estimated from correlation to a whole-embryo RNA-seq time series (22) and measured in minutes after an embryo’s first cell cleavage. (B) Positions of cells from four samples of synchronized embryos on the UMAP plot. (C) Histogram of estimated embryo time for all cells in the dataset. (D) Bar plot showing for bins of embryo time, the percentage of cells in that embryo time bin that we were able to assign to a terminal cell type or pre-terminal lineage. (E) Scatter plot showing correlation of the number of cells of a given anatomical cell class in a single embryo (X axis, log scale) with the number of cells recovered in our data (Y axis, log scale). Each point corresponds to a cell class. Only cells with estimated embryo time >= 390 minutes are included in the counts (many earlier cells are still dividing). Red line is a linear fit, excluding points with y = 0.

We clustered cells in the UMAP using the Louvain algorithm (26) and annotated clusters with cell type identities using marker genes from the literature on C. elegans gene expression (27). Markers used for each annotation are listed in Table S1. The global UMAP arranges cells into a central group of progenitor cells and branches corresponding to eight major tissues (Figs. 1A, S3): muscle/mesoderm, epidermis, pharynx, ciliated neurons, non-ciliated neurons, glia/excretory cells, intestine, and germline. While some individual cell types were identifiable in this global UMAP, many were not, especially progenitor lineages. To gain resolution, we hierarchically created separate UMAPs of each tissue (Figs. S4–S13). These “sub-UMAPs” better resolved specific cell types, allowing us to make extensive, fine-grained annotations.

A combination of marker genes, lineage assignments, and developmental time allowed us to locate 112 specific terminal anatomical cell types, including every lineage input to body wall muscle, every distinct subtype of pharyngeal muscle (pm1–2, pm3–5, pm6, pm7, and pm8) and hypodermis (hyp1–2, hyp3, hyp4–6, hyp7, hyp8–11, seam, and P cells), and every non-neuronal cell type in the mesoderm. We identified 69 of 82 non-pharyngeal neuron types and 9 of 12 glial cell types present in the embryo (Table S2). We could not identify 12 of 14 pharyngeal neuron types. A cluster corresponding to the most differentiated pm3–5 pharyngeal muscle cells had a low level of expression of neuron-specific genes, suggesting that we failed to dissociate the neurons that innervate these muscles in late embryos.

We successfully annotated 93% of cells in our dataset with a cell type (for terminal cells) or a cell lineage (for progenitor cells, discussed below) (Fig. 1D). The number of cells annotated for each cell type was variable but roughly fit the expectation on the basis of the number of cells of that type present in a single embryo (Fig. 1E, r = 0.64, p = 2.4e-13, t test).

Annotation of progenitor lineages

The structure of the global and single-tissue UMAPs was dominated by trajectories of terminal cell differentiation. We hypothesized that closely related lineages could be better resolved by separately analyzing progenitor cells prior to terminal differentiation. Thus, we ran UMAP with only cells with embryo time <= 150, 250, or 300 minutes and found branching patterns that reflect lineage identities (Figs. 2, S14–16). Intestine and germline cells commit to their terminal fates very early and have very divergent expression that distorts the projections, so they were removed and analyzed separately (Figs. S7, S12). The 300-minute UMAP contained several large quasi-connected groups corresponding approximately to major founding lineages, roughly organized by the major fates produced by each founder cell lineage (MS muscle, MS pharynx, C/D muscle and AB-derived lineages that produce either pharynx, neurons/glia, or hypodermis). We were able to resolve additional details by recursively making sub-UMAP projections of these cell subsets.

Fig 2. Annotation of the early lineage.

(A) Diagram showing the position of early mesoderm (MS lineage) cells marked by expression of ceh-51. The lineage radiograph shows the average fluorescent intensity (log10 scaled) of a CEH-51::GFP protein fusion measured by live imaging. The inner rings show the generation of the founder cells, AB (which produces almost exclusively ectoderm and pharynx), MS (mesoderm and pharynx), C (muscle and ectoderm) and P3, which gives rise to P4 (germline) and D (muscle). Daughter cells are named by their relative positions at mitosis (e.g. ABa is the anterior daughter of AB, ABal is left daughter of ABa). (B) UMAP projection of 926 early-stage cells (estimated embryo time <= 150 minutes), colored by embryo time. E lineage and germline cells are excluded and shown separately in Figs S7 and S12, as they differentiate early compared to other lineages. (C) Same UMAP as (B), colored by ceh-51 expression (red indicates cells with >0 UMIs for ceh-51). (D) Expression of hnd-1 and pha-4 measured by sc-RNA-seq (UMAP) and live imaging of GFP protein fusions (radiograph). (E) Cropped section of a UMAP of 8,083 neuron/glia/rectal progenitor cells with embryo time <= 250 minutes (Fig. S15). This plot shows the section of that UMAP that corresponds to the 3,233 cells from the ABpxp ectodermal lineage (“ABpxp” is short-hand for two symmetric lineages, ABplp and ABprp). Colored bold annotations highlight specific lineages that are discussed in the text. (F) Lineage tree for the ABpxppp sub-lineage, highlighting cells that are present in the circled section of (E). The (co-)expression pattern of marker genes identifies branches in the UMAP that correspond to specific ABpxppp descendants. Additional ABpxppp descendants not shown in this panel are annotated in (E), below the circled section.

To annotate progenitor lineages, we exploited lineage marker genes from the literature and the EPiC database, which contains single cell resolution expression profiles extracted by cell tracking software from confocal movies of C. elegans embryos expressing fluorescent reporters (17). In addition to the 180 previously described patterns (17, 28), we have collected movies for 71 additional genes, increasing the total number of patterns in EPiC to 251 genes (Table S3). We annotated branches with lineage identities between the 28-cell and 350-cell stages by finding genes that were differentially expressed both between sister lineages in the EPiC data and between branches of the sub-UMAP trajectories in a concordant manner (Figs. 2, S14–16, Tables S4–S5). For example, expression of ceh-51 is restricted to the MS (mesoderm-producing) lineage (29), allowing us to label the single group of ceh-51(+) cells in 150-minute UMAP as part of the MS lineage (Fig. 2A, B). Within this lineage, we used expression of pha-4 to annotate the anterior granddaughters of MS (MSaa and MSpa) and hnd-1 to annotate the posterior granddaughters (MSap and MSpp) (Fig. 2C). We applied this same logic iteratively across the different UMAPs and lineage marker genes to annotate each branch with its lineage identity (Table S4).

In most cases, branches in the progenitor lineage UMAPs corresponded directly to sister cells in the lineage (Fig. 2D, E), but some branches were unclear or misleading, and marker gene expression was critical to annotate lineages correctly. For example, ABpxpaaaa and ABpxpaapa are cousin lineages, but appear to branch as sisters in the UMAP trajectory, and the same is true for their sisters (ABpxpaaap and ABpxpaapp) (Fig. 2D). In other cases, such as the ABpxppap lineage (Fig. 2D), marker gene combinations were required to annotate lineages that were not contiguous with their parent or sister lineages in the UMAP. These misleading branches demonstrate the importance of having independent expression or lineage data to correctly interpret trajectories visualized in low-dimensional embeddings of sc-RNA-seq data.

To complete our annotations, we used UMAPs of selected subsets of cells with embryo time <= 350 or 400 minutes to reconstruct trajectories leading from the grandparents and parents of terminal cells to their terminal descendants (Fig. S17). Most terminal cell types were thus identified by two methods: first using marker genes for the differentiated cell type, and second by following UMAP trajectories from the cell’s progenitors. Notably, in all cases, the cell type predictions of these two mostly-independent methods were concordant.

A near-complete atlas of the embryonic transcriptome

In total, we annotated 502 distinct cell lineages. Most lineage annotations correspond to a symmetric pair of cells, with the exception of some terminal cell types in which 3–18 cells converge to a homogenous transcriptomic state and could not be further resolved. Our annotations account for 1,068 out of 1,228 individual branches in the C. elegans embryonic lineage (Fig. S18), excluding the 113 branches that lead to programmed cell death. Combined with the dataset of Tintori et al. (8), which profiles the 1- to 16-cell stages, we now have a near-complete molecular atlas of C. elegans embryogenesis.

The lineages included in our atlas partially overlap with the Tintori et al. dataset (8) at the 16-cell stage. Gene expression profiles for lineages annotated in both datasets were concordant (Fig. S19). Additionally, gene expression profiles for terminal cells in our data were concordant with previously published microarray data (30) (Fig. S20).

In Table S6, we provide a map of anatomical cell names to annotations defined in this study. In Tables S7–8, we provide aggregate gene expression profiles for each terminal cell type (binned by embryo time) and each cell lineage annotation. We use bootstrap resampling to estimate a confidence interval for the expression level of each gene in each cell type. In Tables S9–11, we provide lists of differentially expressed genes between all pairs of sister lineages and all pairs of parent vs. daughter lineages within our annotations. Lastly, we systematically re-annotated our previous data from the L2 stage (2), identifying 118 cell types (over twice as many as reported in the initial publication). Tables S12–14 list marker genes, annotation statistics, and expression profiles for the L2 data.

Bifurcating cell fates and multilineage priming

Developmental trajectories in which a parent cell divides to produce two terminal daughter cells of different cell types are a basic type of cell fate decision. Bifurcations like these are common in neuronal lineages in C. elegans, such as those that produce ciliated neurons. To examine the molecular basis for such developmental decisions, we used recursive UMAP projections of ciliated neurons (Fig. 3A) to identify developmental trajectories for all but one of the 22 ciliated neuron types and their parents, missing only the PHA phasmid neurons. The distinction between neuroblasts and terminal neurons was supported by embryo time estimates consistent with terminal cell division times (31), by the expression patterns of cell cycle associated genes and transcription factors (Fig. 3B), and by the structure of the UMAP projection. A 3D version of the UMAP featured better continuity for several trajectories, including those connecting the ASG-AWA, ADF-AWB, and ASJ-AUA neuroblasts with their daughter cells, as well as the branching of the laterally asymmetric left and right ASE neurons (Fig. S21).

Fig 3. Developmental trajectories of ciliated neurons.

(A) UMAP of 10,740 ciliated neurons and precursors. Colors correspond to cell identity. Text labels indicate terminal cell types. Numbers 1–16 indicate parents of 1 ADE-ADA, 2 CEP-URX 3 PHB-HSN 4 IL1 5 OLL 6 OLQ 7 ASJ-AUA 8 ASE 9 ASI 10 ASK 11 ADF-AWB 12 ASG-AWA 13 ADL 14 ASH-RIB 15 AFD-RMD 16 AWC-SAA (purple) and BAG-SMD (red). 4–6, 8–10, and 13 are listed as parents of only one cell type as the sister cells die. Numbers 17–20 indicate grandparents of 17 IL1 (= IL2 parent) 18 OLQ-URY 19, 20 ASE-ASJ-AUA. Differentiated PHA was not conclusively identified but may co-cluster with PHB. The parent of PHA is not present in this UMAP, but was located separately within the area annotated as “rectal cells” in the UMAP in Fig. S3. The tiny cluster labeled with an asterisk (*) is putatively AWC-ON on the basis of srt-28 expression. (B) UMAP plot colored by embryo time (colors matched to Fig. 1A) and gene expression (red indicates >0 reads for the listed gene). egl-21 codes for an enzyme that is essential for processing neuropeptides (51). Its expression is used as a proxy for the onset of neuron differentiation. mcm-7 codes for a DNA replication licensing factor. Loss of mcm-7 expression in each UMAP trajectory approximately marks the boundary between neuroblasts and terminal cells. unc-130 is known to be expressed in the ASG-AWA neuroblast but neither terminal cell (52). (C) Cartoon illustrating the lineage of the ASE, ASJ, and AUA neurons. (D) Heatmap showing patterns of differential transcription factor expression associated with branches in the ASE-ASJ-AUA lineage. Expression values are log-transformed, then centered and scaled by standard deviation for each row (gene).

To identify potential regulators of cell fate decisions, we identified genes that were differentially expressed between the branches of each bifurcating ciliated neuron lineage (Table S9). The lineage of the ASE, ASJ, and AUA neurons (spanning embryo time ~215–650 minutes) serves as a representative example (Fig. 3C). About 3–4 TFs are specific to each terminal neuron type in this lineage (Fig. 3D). Similar numbers of branch-specific TFs were observed for other lineage bifurcations (Fig. S22). Beyond these simple cases, we also found several TFs that were expressed in a parent cell and had expression selectively maintained in one daughter but not the other. For example, the TFs ceh-36/37/43/45, ham-1, and hlh-3 are all co-expressed within single ASE-ASJ-AUA neuroblast cells. ceh-36/37 and hlh-3 expression was maintained in only one daughter of this neuroblast, the ASE parent, while ceh-43/45 and ham-1 expression was maintained only in the other daughter, the ASJ-AUA neuroblast (Fig. S23).

This pattern, where a progenitor cell co-expresses genes specific to each of its daughters, has been termed “multilineage priming” and has been observed in several organisms and developmental contexts (10, 32–36). Our transcriptomic atlas of the C. elegans cell lineage allows us to provide an unbiased quantification of the prevalence of multilineage priming throughout the organism’s ectoderm and mesoderm (we lack sufficient resolution in our annotations of the endoderm, which produces only one cell type, the intestine). There are 172 instances in which we have data for a parent cell and both of its distinct daughters. Of these, 52% exhibit multilineage priming. Multilineage priming events are distributed throughout several generations of both the ectoderm and mesoderm (Fig. S24), demonstrating that it is a common and pervasive mechanism of gene regulation. The expression patterns of many TFs involved in multilineage priming, e.g. hlh-3 (Fig. S23D), are confirmed by the movies in EPiC (17).

Transcription factors that are both required for neuron type specification and have expression maintained throughout the lifetime of the neuron are referred to as “terminal selectors” (37). To identify potential terminal selectors, we looked for transcription factors that were 1) expressed in a neuron type but not its sister in the embryo and 2) expressed in the same neuron type at the L2 stage. This analysis replicated 23 known neuron-TF associations (37) and identified 116 novel associations (Table S15). Other known associations may have been missed due to the extreme sparsity of the L2 stage data, and the fact that many terminal selectors are expressed at low levels in fully differentiated neurons, or are expressed in both daughters of a terminal division. In cases where a neuron’s sister undergoes programmed cell death, we looked for TFs that are both enriched in the terminal cell’s most recent ancestor that has a surviving sister cell (compared to that sister), and also have expression maintained throughout the lifespan of the terminal neuron. This revealed novel associations, including ceh-6 for AVH, ceh-8 for RIA, unc-62 for RIC, and lin-11 for RIC and RIM, in which the putative terminal selector TF is expressed in a neuroblast before the terminal cell is produced, suggesting that these lineages commit to a cell fate early.

Only two neurons (ASE and AWC) are known to have left-right asymmetric gene expression (38, 39). For both neuron types, the lineages of the left and right neurons diverge in the early embryo at the 4-cell stage (< 50 minutes). Asymmetric gene expression in our data, however, emerges only much later in embryogenesis. The transcriptomes of ASEL and ASER diverged in our UMAP at ~650–700 minutes, with lim-6 expressed specifically in the ASEL branch, consistent with previous studies (40, 41). AWC left/right asymmetry occurs stochastically, with one neuron becoming “AWC-ON” and the other becoming “AWC-OFF” (39). We identified a small cluster in the UMAP with embryo time >700 minutes as AWC-ON based on srt-28 expression (Fig. 3A) (42). AWC-OFF is putatively part of the main AWC trajectory. No evidence of left/right asymmetry was observed in neurons besides ASE and AWC.

Transcriptional convergence of co-fated lineages

While most bilaterally symmetric cells were not distinguishable by UMAP (as expected), several cell types with >2-fold symmetry are produced by multiple non-symmetric lineage inputs. These lineage inputs tended to cluster separately in our progenitor cell UMAPs, while in our late-cell tissue UMAPs, we saw almost no evidence of heterogeneity within the terminal cell types that they produce. This difference suggested that the transcriptomes of these co-fated lineages were converging during differentiation.

One example of apparent molecular convergence of cells from distinct lineages was the IL1-IL2 neuroblasts. The six IL1 and six IL2 neurons are produced by three symmetric pairs of neuroblast lineages. Each neuroblast pair produces a pair of bilaterally symmetric IL1 neurons, and likewise a pair of IL2 neurons. A UMAP of IL1/2 neurons and progenitors revealed trajectories for these neuroblasts that converge gradually over their lifespan (Fig. 4A). The transcription factor ast-1 was transiently expressed at extremely high levels (>10,000 TPM) during this process, suggesting that it might play a role in homogenizing the IL1/2 neuroblast transcriptomes (Fig. 4B). Correspondingly, expression of genes differentially expressed between the input lineages decreased over time, while expression of genes specific to terminal neurons increased (Fig. 4C–D). We observed similar lineage convergence via continuous gene expression trajectories for other cell types, including hypodermis (Fig. S8), head body wall muscle (Fig. S17), and GLR cells (Fig. S17).

Fig 4. Full vs. incomplete convergence of lineages producing common cell types.

(A) UMAP of 854 IL1/2 neurons and progenitors colored by estimated embryo time (cells selected on the basis of annotations in Figs. 3A and S15). (B) IL1/2 UMAP colored by ast-1 expression level (log2 size-factor normalized UMI counts). (C) IL1/2 UMAP colored by expression of unc-39, a gene specific to branch 1. (D) Heatmap showing the average expression level of lineage specific and terminal cell type specific genes over time for each of the 3 branches. (E) Fig. S5A shows a UMAP of body wall muscle and mesoderm cells. This panel is a zoomed-in view of that UMAP, including only 17,520 BWM cells, which are grouped into “bands” based on marker gene expression patterns (here, a cell is considered to express a gene if it or >= 2 of 5 of its nearest neighbors have >0 reads for the gene). (F) Physical positions of cells in each BWM band (colors matched to panel E) in the embryo at 430 minutes. Adapted from Fig. 8B of (16). (G) Transcriptome Jensen-Shannon distance for posterior (orange+green bands in panel E) BWM vs. row 2 (blue band) or row 1 (pink band) head BWM over time. Heterogeneity between BWM subsets persists throughout development and may reflect functional differences.

Like the IL1/2 neurons, IL socket glia (ILso) are produced by three symmetric pairs of lineages. In contrast to the examples discussed above, trajectories formed by the ILso progenitors and their terminal descendants were discontinuous in UMAP space (Fig. S25). Discontinuous trajectories were also observed for several other cell types from multiple tissues, including other glia, several neuron types, the excretory gland, coelomocytes, and somatic gonad precursors (Z1/Z4) (Fig. S25). Several lines of evidence suggest that these discontinuities reflect sudden changes in the transcriptome rather than technical artifacts of sc-RNA-seq or UMAP. Discontinuous trajectories had more genes differentially expressed between the parent and daughter cells than continuous trajectories (Fig. S26). Almost all discontinuous trajectories were observed in lineages where a parent cell gives rise to two daughters of different broadly-defined cell types, e.g. a glia and a non-glial cell, or a ciliated neuron and a non-ciliated neuron (Fig. S26). These discontinuities were seen in both the global and the tissue-specific UMAPs, and with different UMAP parameters. Finally, for most discontinuous trajectories, cells had a continuous distribution of embryo times (Fig. S27). However, a few trajectories, such as that of the BAG neuron, had gaps in the embryo time distribution indicative of potential sampling bias.

Body wall muscle (BWM) was exceptional in that lineage-related heterogeneity persisted throughout differentiation. BWM is produced by multiple distinct lineages (C, D, MS) and occupies a wide range of positions along the anterior-posterior (A-P) axis of the animal. A UMAP of BWM cells identified distinct trajectories for the 1^st row of head BWM vs. all other BWM (Fig. 4E). The non-1^st-row trajectory was formed by input trajectories that corresponded to lineages and progressed in parallel along the temporal axis. Using marker genes that are expressed in domains along the A-P axis (17, 43–45), we divided BWM cells in the UMAP into six “bands” (Fig. 4E) and identified the specific anatomical cells present in each band (Fig. 4F, Table S16). We found that the Jensen-Shannon (JS) distance, a measure of transcriptome difference, between the transcriptomes of posterior BWM (C lineage) vs. both the 1^st and 2^nd rows of BWM (D/MS lineage) did not decrease over time (Fig. 4G), indicating that BWM heterogeneity persists throughout differentiation.

Temporal dynamics of the lineage-transcriptome relationship

The presence of discontinuities between progenitor cells and terminal cells in the UMAP projections suggested that the terminal division could mark a shift from lineage-correlated to fate-correlated gene expression. We asked how well the distance between two cells in the lineage predicts the difference between their transcriptomes (as defined with the JS distance). We focused on the AB lineage, which produces mostly ectoderm and accounts for ~70% of the terminal cells in the embryo. The AB lineage undergoes roughly synchronized cell divisions, allowing us to group cells by generation. For example, we refer to the 32 cells produced by 5 divisions of AB as “AB5” and so on.

In AB5 (early/mid-gastrulation; 50-cell stage), the earliest stage where our lineage annotations were near-complete, sister cells were more similar than distant relatives, but the difference was not large (Fig. 5A). In AB6 (mid-gastrulation; 100-cell stage) and AB7 (late gastrulation; 200-cell stage), the transcriptomes of sister cells become more similar than in AB5, while those of distant relatives become more divergent, resulting in a strong correlation between transcriptome distance and lineage distance. In AB8 (350-cell stage), most epidermal cells exit the cell cycle and begin terminal differentiation, while neuron/glia progenitors continue for 1–2 more cell divisions. AB8 thus features a bimodal distribution of transcriptome JS distances: terminal epidermal cells become highly distinct from neuron/glia progenitors, but cells within each group are more similar (Fig. S28). Finally, most neuron/glia progenitors in AB8 produce two terminal daughters in AB9 that have distinct cell fates and a much weaker lineage-transcriptome correlation than in earlier generations.

Fig 5. Correlation between cell lineage and the transcriptome in the ectoderm.

(A) Jensen-Shannon (JS) distance between the transcriptomes of pairs of ectodermal cells (AB lineage), faceted by cell generation and lineage distance. AB5 refers to the cell generation produced by 5 divisions of the AB founder cell, and likewise for generations AB6–9. The “transcriptome” of a given anatomical cell is defined as the average gene expression profile of all sc-RNA-seq cells annotated as that anatomical cell. Pairs of bilaterally symmetric cells are excluded from the statistics. (B) Estimates of the extent to which lineage predicts the transcriptome in AB5–9. (C) Distribution of the number of “lineage signature transcription factors”—TFs that distinguish a cell from its sister—for all cells in AB5–9. The outlier points in AB8 are instances where a terminal epidermal cell is a sister of a neuroblast. (D) Proportion of lineage signature transcription factors for a cell in a given generation that have expression maintained in 0, 1, or 2 of the cell’s daughters in the subsequent generation. (E) Proportion of lineage signature TFs for which expression in a given cell was maintained from the cell’s parent vs. newly activated after the parent’s division.

Together, these statistics suggest that progenitor cells develop strong expression signatures of their lineage identity, and that these signatures are rapidly lost or overshadowed by new expression at the time of the terminal division. An analysis of cells from the mesoderm (MS lineage) replicated the trends observed in the ectoderm (Fig. S29A).

To summarize the strength of the lineage-transcriptome correlation in a cell generation as a single number, we developed a statistic analogous to the concept of pseudo-R² in generalized linear regression models (46). Consistent with the above analysis, we find that the extent to which lineage predicts the transcriptome increases throughout gastrulation, peaks at 55% in AB7, and then falls to 18% after terminal differentiation in AB9 (Fig. 5B). Next, we asked how much of the total pseudo-R² for one cell generation was attributable to gene expression signatures associated with each preceding cell generation. For cells in AB5–8, the largest contributor to pseudo-R² was the identity of their ancestor in the AB3 generation (Fig. S30). This is interesting because many of the clades formed at AB3 share a broadly-defined tissue fate. For example, the clade founded by the cell ABala produces only neurons and glia, while the clade founded by the cell ABarp produces mostly (but not exclusively) epidermal cells. The second largest lineage signal was from the identity of a cell’s parent in the preceding generation (i.e. the tendency of sister cells to be more similar than cousins). Thus, both broad and fine-grained structure in the lineage contribute towards shaping the transcriptome.

To investigate the potential regulatory mechanisms that differentiate sister cells, we identified transcription factors (TFs) that distinguish each cell in AB5–9 from its sister. The median number of these “lineage signature TFs” per cell increased over time, ranging from 1.5 in AB5 to 14 in AB9 (Fig. 5C). A substantial number of lineage signature TFs (~40–50%) had expression selectively maintained in only one of a cell’s two daughters (Fig. 5D). In other words, TFs that distinguish a cell from its sister in one generation are frequently re-used to distinguish that cell’s daughters from each other. Sister cells are also differentiated by the expression of new TFs not present in their parents. The proportion of lineage signature TFs that are newly expressed ranged from 33–61% and increased over time in AB6–9 (Fig. 5E). Temporal dynamics of lineage signature TFs were similar in the mesoderm (Fig. S29).

Taken together, these results highlight the incremental nature of cell fate decisions: every terminal cell is the result of a series of lineage bifurcations, each of which, on average, involves multiple differentially expressed TFs.

Global patterns of gene expression and transcriptome specialization

Hierarchical clustering of expression levels in all annotated lineages and cell types (Tables S7–8) provides a global view of expression dynamics for all genes in our dataset. A heatmap of pre-terminal lineage expression profiles (Fig. S31) does not reveal large clusters of genes specific to specific lineages, other than one cluster of genes specific to the early C and D lineages. Similarly, most marker genes used for lineage annotation are not part of large clusters of co-expressed genes. The clusters that do form are composed of early tissue-specific genes. The lack of cluster structure in the heatmap suggests that differential fates for tissue sub-lineages are specified by relatively small sets of genes. By contrast, a heatmap of terminal cell type expression profiles (Fig. S32) has more obvious structure. Cells in each major tissue express ~500–1500 tissue-enriched genes. There is little reuse of tissue-enriched genes between tissues other than hypodermis, which shares many genes with glia and intestine. Neuron subtypes and other specialized cells (such as the hmc or M cell) are typically distinguished from other cells within their tissue by expression of <20–300 genes. Finally, there are substantial temporal changes in expression, especially in muscle and hypodermis.

We observed substantial variation between cells in the Gini coefficient, which measures how unequally different genes are expressed in a given cell type (Fig. S33A). Hypodermis, seam cells, and the pharyngeal gland express small sets of cell type specific genes at very high levels (high Gini coefficient), while the intestine and germline feature diverse gene expression patterns (low Gini coefficient). In several cell types, such as the pharyngeal gland, increases in Gini coefficient over time coincide with decreases in the number of TFs expressed per cell (Fig. S33B). Families of TFs also exhibit differential expression patterns over time and across lineages. Nuclear hormone receptors (NHRs) are on average activated later in development than other TF families, such as Forkhead and Homeodomain TFs (Fig. S33C). Hypodermis and intestine express many distinct NHRs, while expression of Sox family TFs is largely restricted to neurons, glia and pharynx (Fig. S33D).

An RShiny app to explore and extend on our analysis

We developed VisCello to distribute single cell analyses and provide interactive visualizations (Fig. S34). It is available as a web app (https://cello.shinyapps.io/celegans/) and can also be installed as an R package (https://github.com/qinzhu/VisCello.celegans). VisCello hosts dimensionality reductions (e.g. UMAPs), cell annotations, and marker gene tables for the different subsets of the data described in this manuscript. Users can visualize gene expression on UMAP or PCA plots, on a lineage tree diagram, or as box/violin plots grouped by cell type or lineage. The plots are interactive, allowing users to zoom in on subsets of cells, define new cell annotation groups, and run differential expression analysis and GO/KEGG enrichment with these newly defined groups. Program state can be downloaded and shared, facilitating collaboration. VisCello can also be used to host and disseminate other single cell datasets, including data from the C. elegans 1–16 cell stage (8) and L2 stage (2) (https://github.com/qinzhu/VisCello).

Discussion

The cells of C. elegans are limited in number and invariant in lineage and cell fate, making it feasible to conduct comprehensive, whole-organism investigations. Yet within this limited repertoire of cells exists an impressive diversity of cell types, which work together to produce complex anatomical structures and behaviors. This study and our previous work (2, 17) have shed light on the molecular basis for the specification of these cell types, but are only the first step toward a comprehensive understanding of the molecular basis of development. We hope that this resource will help guide future projects in the C. elegans community.

In contrast to developmental sc-RNA-seq datasets from other species, this dataset links gene expression trajectories to the exact cell lineages they correspond to, allowing steps in the process of differentiation to be associated with specific cell division events. Thus, our data provide a quantitative portrait of Waddington’s landscape for a whole organism. The abruptness of many cell fate decisions in C. elegans, with many distinct terminal cell types becoming distinguished only in the final embryonic cell division, contrasts, however, with the smooth landscape in Waddington’s illustrations and warrants further investigation.

We observe convergence of gene expression patterns in many instances where distinct cell lineages produce identical or related cell types. Data from a recent atlas of mouse organogenesis (13) suggests that this phenomenon is also prevalent in vertebrates. For example, myocytes in the mouse atlas are produced by two convergent trajectories, and excitatory neurons are produced by several trajectories.

Our analysis highlights two important challenges that will be faced by efforts to reconstruct the cell lineages of other organisms using single cell RNA-seq. First is the difficulty of accurately connecting developmental trajectories that start after the convergence of lineages with similar cell fates to trajectories that span earlier stages of development. A naive interpretation of the UMAP projection of the full dataset (Fig. 1A) could lead to inferred trajectories that are inconsistent with the correct lineage (for example, incorrectly concluding that hypodermis and seam cells are produced from a common ancestor that previously diverged from the progenitors of neurons). Second is the difficulty of constructing continuous trajectories for lineages that undergo abrupt changes in gene expression. In our data, progenitor cells that give rise to glia, excretory cells, and non-ciliated neurons were more often than not disconnected to their terminal daughters in UMAP space (Figs. S25–26), reflecting the fact that many of these lineages only commit to a terminal fate after their final cell division.

Due to these challenges, we anticipate that constructing end-to-end trajectories of vertebrate organogenesis will require single cell RNA-seq to be integrated with experimental lineage tracing methods (47). It will also require improved computational methods that can model heterogeneity among poorly-differentiated progenitor cells and highly-differentiated cell types in an integrated manner.

Between this study, our previous study of the L2 stage (2), and earlier studies of the 1 to 16-cell stage embryos (19, 20), a large portion of the early C. elegans life-cycle has now been profiled by single cell transcriptomics. However, more datasets will be needed to complete missing stages, including other larval stages and the adult soma and germline. In the future, single cell profiling of different strains or species will be a useful approach to examine the evolution of cell types and their expression programs. All of these datasets will ideally be integrated into a single visualization platform, such as VisCello, to allow full tracking of cell trajectories from fertilization through the end of life. A greater challenge will be to discover the precise mechanisms that produce transcriptomic outputs. Single cell transcriptome analysis of mutants will likely need to be integrated with new single cell multi-omic technologies (48) to bring mechanistic studies to a whole-organism scale.

Methods Synopsis

Sample preparation:

C. elegans embryos were prepared by standard hypochlorite treatment methods from populations of synchronized early adult worms grown at 20° C. Embryos were dissociated immediately or aged in egg buffer prior to dissociation. Eggshells were removed by chitinase digestion, embryos were dissociated by manual shearing, and single cells were isolated by filtration or centrifugation. Single cell RNA-seq libraries were generated with 10X Genomics v2 chemistry and standard protocols, and sequenced on Illumina NextSeq instruments.

Computational analysis:

The single cell RNA-seq data was processed with the 10X Genomics CellRanger pipeline, aligning reads to a modified version of the WormBase (27) WS260 reference transcriptome that had transcript 3’ UTRs extended by 0–500 base pairs. The data was then visualized using dimensionality reduction methods. Single cell transcriptomes were first projected into 50–100 dimensions (depending on the analysis) using principal components analysis, and then projected into 2 or 3 dimensions using the UMAP algorithm (23). Cells in the UMAP space were clustered using the Louvain algorithm (26). For each cell, the age of the embryo that it came from (“embryo time”) was estimated by correlating its transcriptome with a bulk RNA-seq time series (22). Cells were then manually annotated with their corresponding cell type and lineage, as described in the main text. The annotation process was guided by the UMAP projections, Louvain clusters, and embryo time estimates, but also relied heavily on fluorescent reporter imaging data from the EPiC (17) and WormBase (27) databases.