Lineage-resolved analysis of embryonic gene expression evolution in C. elegans and C. briggsae

Abstract

The constraints that govern the evolution of gene expression patterns across development remain unclear. Single-cell RNA-sequencing can detail these constraints by systematically profiling homologous cells. The conserved invariant embryonic lineage of C. elegans and C. briggsae makes them ideal for comparing cell type gene expression across evolution. Measuring the spatiotemporal divergence of gene expression across embryogenesis, we find a high level of similarity in gene expression programs between species despite tens of millions of years of evolutionary divergence. Nonetheless, thousands of genes show divergence in their cell-type specific expression patterns, with enrichment for functions in environmental response and behavior. Neuronal cell types show higher divergence than others such as the intestine and germline. This work identifies likely constraints on the evolution of developmental gene expression.

Introduction

How the patterns of developmental gene expression, from progenitor to terminal cell types, change across evolution remains unknown. Recent single-cell developmental genomics studies have demonstrated the possibility of comprehensively measuring the transcriptomes of all of the diverse progenitor and terminal cell types present in animal embryos. Changes in gene regulation during development can have adaptive function, or could occur despite apparently conserved cellular function, a phenomenon termed developmental systems drift. Because most prior studies focused on small numbers of genes or cell types, it is not known the extent to which gene expression and its dynamics change across evolution for the full set of genes and cell types.

Rationale

Caenorhabditis elegans is a widely studied model organism for development, notable for its invariant lineage and complete cell type catalogue. The related species C. briggsae is over 20 million years diverged from C. elegans, yet they develop through an essentially identical lineage of developmental progenitors resulting in the same list of terminal cell types. This makes Caenorhabditis spp an ideal system to compare gene expression for equivalent cell types across evolution to learn for which cells and genes have more or less expression divergence.

Results

We compared gene expression in over 175,000 cells for each species covering embryogenesis from gastrulation through terminal differentiation. We annotated these cells to identify 429 shared progenitor and terminal cell types, allowing comparisons of gene expression at full-organism scale. We then identified features predicting conserved expression for each gene and cell type. We found that most gene expression is conserved, but expression of thousands of genes has diverged between these species, including a substantial number that have undergone changes in expression timing. In particular, the expression of developmental transcription factors and their predicted gene regulatory networks appeared largely conserved between species. Comparing homologous cell types between species we found that some cell types, including the embryonic neurons, showed higher rates of transcriptome evolution than others, such as intestine and muscle. Across developmental time, the midpoint of embryogenesis on average had reduced transcriptomic change, providing additional evidence for the postulated ‘developmental hourglass’ model. However, these patterns of stage-specific conservation varied between cell types. Finally, we determined the evolutionary expression fates of many duplicated genes, including cases of possible neo- and sub-functionalization and pseudogenization.

Conclusions

This work provides a lineage-resolved understanding of genome-wide expression evolution at whole-embryo scale. We provide quantitative measurements of the extent of gene expression evolution and infer cell specific variation in gene expression. The observed expression differences provide numerous examples that can guide future mechanistic studies of regulatory evolution. Finally, this dataset will provide a comprehensive resource to guide the study of gene expression evolution in systems where the lineage conservation is not known.

Graphical Abstract

Fig. 0. The transcriptomes of single cells from C. elegans and C. briggsae were collected, annotated, and compared to identify differences in expression conservation for cell types and genes. The lineage-resolved transcriptomic atlases were used to measure the conservation of gene regulatory networks, contrast the timing of gene expression between species, and classify gene duplications based on expression and protein conservation.

Introduction

Animals can evolutionarily adapt to changing environments while maintaining core functions, in part through modifications to developmental gene expression profiles of individual cell types. The comparison of single cell transcriptomes of divergent species can reveal developmental and functional patterns of gene expression evolution, including those following gene duplications (1–11). Such changes could reflect adaptations to new environments or phenomena such as “Developmental Systems Drift” (12–14). The ability to profile the transcriptomes of all the individual cell types across species provides an opportunity to determine the developmental evolution of gene expression and cell type variation.

The nematodes C. elegans and C. briggsae provide an ideal platform for investigating these questions. C. elegans embryos produce 558 anatomically-defined terminal cells via a fully described sequence of invariant cell divisions (15–17). Embryonic gene expression in C. elegans is well characterized, with several single-cell resolution imaging and RNA-sequencing studies having revealed the expression profiles of nearly all progenitor lineage states and embryonic terminal cell types (18–22). C. briggsae is more than 20 million years divergent with 1.78 substitutions per neutral site, yet has essentially the same lineage and terminal cell types (16, 17), and is nearly identical in morphology, providing an opportunity to study molecular process divergence under developmental constancy (23, 24).

Here, we determined the gene expression of progenitor and terminal cell types in C. elegans and C. briggsae embryos by single-cell RNA sequencing of >175,000 cells per species. Using the enumerated list of cell types and independently determined cell markers to validate the completeness and accuracy of annotations, we defined ~429 shared progenitor and terminal cell types (>150x coverage per cell type in each species). The data reveal the variation of gene expression across cell types and the changes in gene usage of the different cell types.

Results

Aligned single-cell transcriptome atlases of C. elegans and C. briggsae

We measured mRNA levels of protein coding genes for embryonic cells from the C. briggsae by single-cell RNA-seq, using the 10x Genomics platform. We sampled a time course of temporally overlapping synchronized embryo populations with the goal of maximizing representation of embryonic stages from gastrulation (~28-cell stage) through terminal differentiation. After removing doublets and low-quality cells, the C. briggsae dataset consisted of 178,818 cells from seven independent cell preparations (~145x average coverage of the 558 terminal cells and 670 progenitors branches in the shared C. elegans/C. briggsae lineage). We compared these cells to a pre-existing C. elegans single-cell embryonic atlas (18), supplemented with 25,011 newly-collected wild type cells from late-stage embryos. We also included an additional 104,549 cells from three C. elegans transcription factor mutants that only affect very specific cell types, filtered to only include cell types lacking expression of that transcription factor in wild type (see materials and methods). After removing doublets and low-quality cells, these samples yielded a total of 210,937 C. elegans cells (~172x coverage).

Both the C. elegans and C. briggsae genomes contain ~20,000 genes. We identified a set of 13,679 orthologs, which we refer to for simplicity as the 1:1 genes. This set combined previous orthology sets (11,133 genes) with additional gene pairs (2,546 genes) identified by gene synteny, reciprocal BLAST searches, and pairwise alignments (Fig. S1, see materials and methods). We also classified genes with more complex orthology relationships (such as 1:many, many:many, 1:0) into orthogroups ((25, 26); Table S1) to assess expression conservation of these genes.

While the embryonic lineage trees are nearly perfectly conserved, it was unclear if the extent of molecular identity conservation of each cell type would be sufficient to allow co-identification of homologous cell types of the two species. In a joint transcriptional space using the 1:1 genes, the cells from the two species clustered well, allowing us to identify shared cell types. (Fig. 1A–C; see materials and methods). For the C. elegans cells these annotations agreed well with their prior annotations from (18) (Table S2). We observed similar proportions of cells from C. elegans and C. briggsae in each progenitor and terminal cell type annotation, which correspond to 496 out of the 558 terminal cell types and the parents of 54 additional, mostly late-born terminal cell types. The C. briggsae samples were slightly enriched for earlier cells since this dataset included more early-staged samples (Fig. 1D; Fig. S2). Independent estimates of embryo stage based on comparisons to a whole-embryo time course agreed with the expectation from the staging (Fig. S2) and birth times in the lineage (Fig. S3). A few cell types were detected only in one species (i.e. <10% of cells in the cluster were from C. briggsae). These relative depletions could reflect a small number of true differences in cell type, but more likely reflect biases in their amenability to dissociation by the methods used here (e.g. pharyngeal neurons) or a lack of conserved cell type markers (see Table S3). These cells were not considered in subsequent comparative analyses.

Fig. 1. Homology in annotation of cell types between C. elegans and C. briggsae.

(A, B) Using the orthologous genes between the species, the datasets were co-embedded in a shared reduced dimension space. The terminal and progenitor cell type annotations for C. elegans (A) and C. briggsae (B) are shown with color labeling matching cell type identities in the two species. (C) Overview of which cell types were captured and annotated in C. elegans and/or C. briggsae, shown as a circular plot of the developmental lineages, highlighting the breadth of the dataset coverage. (D) The progenitor cell types (shown by their originating division stage) and terminal cell types (shown by their cell class) as percentages of total number of cells indicates the similarity of the datasets, with a slight bias towards cells from younger embryos in C. briggsae.

Comparing our observations to previous studies investigating the expression pattern of individual homologous genes between C. elegans and C. briggsae, we found agreement for 9 of 10 genes, supporting the accuracy of the cell type annotations (27–38); Table S4). The high concordance of marker expression across species indicates that the conserved lineage and anatomy in these two species reflect recognizable, homologous, molecular states. This annotation provides a basis for comparative analysis of homologous cells and genes across development.

Gene-level constraints on expression patterns

We compared the expression patterns of each 1:1 gene across cell types between the two species, measuring where a gene is expressed, the relative levels of a gene between cell types, and the overall expression level. Many genes with known cell-type specific expression in C. elegans have conserved expression patterns and levels in C. briggsae. For example, pha-4/FoxA, a master regulator of pharyngeal fate, is consistently expressed in all pharyngeal cell types in both species (Fig. 2A; Fig. S4; (39, 40)) and hlh-4/Achaete-Scute has conserved, high expression exclusively in the ADL neuron in both species (Fig. S5). Similarly, many genes, such as the translation elongation gene, eef-2 (Fig. S6), were consistently broadly or ubiquitously expressed in both species. For other genes, we observed substantial changes in the absolute value of expression across cell types despite the similarity of expression patterns. For example, the Rab GTPase, rab-7 was detected in all cells of both species but at 3-fold higher levels in C. briggsae than in C. elegans (Fig. 2B; Fig. S7). Distinguishing biological from technical explanations for genes with global quantitative differences is challenging, so we didn’t analyze this group further, but the data provide candidates for future study. Some genes had completely different patterns in the two species, such as the transmembrane gene, T04A6.1, which is exclusively expressed in the neurons ADL, ADF and ASH in C. briggsae and in a different neuron, RMG, in C. elegans (Fig. 2C; Fig. S8). Conversely, other genes had only partial overlap in the identity of expressing cells. For example, the homeodomain gene vab-15 is expressed in both species in the neurons PVP, PVT and AVG and a few other cell types, but is expressed in ABp(l/r)pa-derived (mostly neurogenic progenitor) sublineages in C. briggsae but not C. elegans (Fig. 2D; Fig. S9).

Fig. 2. Orthologous genes between C. elegans and C. briggsae are mostly conserved in their gene expression patterns across embryonic development.

(A) The gene expression of pha-4, (B) rab-7, and (C) T04A6.1 for C. elegans and C. briggsae show a diversity of conservation patterns across the terminal cell types. Expression values are shown for cell types with confident expression (having greater than zero expression in the lower bound of a 95% confidence interval and greater than 80 TPM in either species). (D) The gene expression of vab-15 for C. elegans and C. briggsae shows divergence in expression in the ABp(l/r)pa-derived sub-lineages (highlighted in red). (E) The distribution of JSD_gene values for all well-detected genes (>80 TPM in at least one cell type) indicates that most genes have a conserved expression pattern. Each example gene is shown with a 95% confidence interval (red box) with the median value displayed as a dark red line. (F) Comparison between the JSD_gene calculated on expression in the progenitors and terminal cell types for all genes that are well detected in both the progenitor or the terminal cell types.

To evaluate expression pattern conservation of each gene between C. elegans and C. briggsae, we used the Jensen-Shannon Distance (JSD_gene), calculated on the expression levels (Transcript per Million (TPM)) in all homologous cell types, herein called the ‘gene distance’ (Table S5, see Fig. S10 for comparison of JSD_gene to other metrics). This metric measures distance between two frequency distributions over discrete classes in terms of how well one distribution (C. elegans) predicts the second distribution (C. briggsae). Gene distance values range between 0 and 1, with lower values indicating less difference between the distributions and more conserved expression. We considered genes that were confidently detected (TPM > 80 in at least one cell type) in both species, excluding lowly-expressed genes that may represent real differences but for which it is difficult to rule out technical artifacts. Most of these genes had clearly conserved expression (5,500 genes with JSD_gene < 0.45 out of 9,954 expressed genes), while a smaller number differed in expression pattern (1,849 genes with JSD_gene > 0.65 out of 9,954; Fig. 2E). The fact that both rab-7 and eef-2 have low distances, despite rab-7 differing between species in overall level, highlights that this metric captures divergence in pattern, not level.

We compared gene expression pattern conservation between terminal and progenitor cell types (Fig. 2F). A majority of genes expressed in both species showed good agreement in their gene distance between terminal and progenitor cell types. However, a subset of genes showed conserved expression patterns in terminal cell types, with more divergent patterns in the progenitor cell types. Most of these genes had greater expression in the terminal cell types. By contrast, there were few examples of genes with similar expression in progenitors and divergent expression in terminal cells, possibly due to the smaller number of genes detected at higher levels in progenitors but not terminal cells.

Gene expression breadth remains consistent between species

We next assessed how well the breadth of expression is conserved. To measure the cell-type specificity of expression for each gene, we calculated Tau (41), which considers distribution evenness over discrete classes. Tau varies between zero and one, with a low value corresponding to broad expression, and a high value indicating patterned or cell type specific expression. Tau shows a bimodal distribution in each species, peaking in the lower range with many broadly-expressed genes, such as rab-7 and eef-2, and in the upper range with specifically-expressed genes (hlh-4, vab-15, and T04A6.1; Fig. 3A). Between these two peaks were genes corresponding to expression that was not ubiquitous, but present in multiple cell types (e.g. pha-4). To facilitate downstream analyses, we used Tau to classify genes as “broad” (Tau < 0.4), “patterned” (0.4 < Tau < 0.7) or “specific” (Tau > 0.7). Tau was highly conserved between species (R² = 0.77), indicating that genes generally do not rapidly change in their expression breadth. Only 153 genes expressed in both species had a difference in Tau of greater than 0.3. Many of these appear to reflect differences in the temporal persistence of early expressed genes (107 genes). For example, in C. elegans, transcripts for the RNA binding protein gld-1 and the p53 homolog cep-1 are expressed in the germline, maternally provided to the embryo and disappear rapidly in zygotic lineages with no expression in terminal somatic cells (Fig. S11; (20)). In contrast, transcripts for these genes in C. briggsae remain present throughout cleavage and even in some terminal somatic cells, suggesting increased stability or zygotic re-expression. A disproportionate fraction of genes with different Tau values between species showed persistence in C. briggsae while fewer showed similar patterns in C. elegans (71 genes in C. briggsae and 36 in C. elegans), suggesting a global difference in dynamics of some maternal transcripts between the species (Table S6). Overall, these observations suggest that it is rare for genes to transition between broad and specific patterns of expression during evolution, but also implicates some flexibility in the temporal domain of expression.

Fig. 3. Intersection of gene expression breadth and conservation reveals gene family level differences.

(A) The Tau values from either species for well detected genes, compared with the maximum gene expression of either species. (B) The mean of the Tau values from either species, calculated on the terminal cell types compared to the terminal JSD_gene with the maximum gene expression. (C) A Sankey plot, showing the count and proportion of genes falling into the listed categories with a TPM > 80 in at least one species. (D) The JSD_gene for genes with a C. elegans embryonic RNAi or allele lethal phenotype is lower by a Bonferroni corrected Wilcoxon Rank Sum test (all pairwise comparisons showed significance except embryonic lethal to larval specific lethal). (E) Fold enrichment of the WormCat categories within the conservation and gene expression breadth categories for the progenitor and terminal cell types. The significance of the enrichment is from a Bonferroni corrected Fisher’s exact test (* is p <0.05, ** is p <0.005, *** is p <0.0005). (F) The expression of members of the cilia gene program, shown as the max normalized values (the gene expression in TPM divided by the maximum TPM of that gene in that species) in the ciliated neurons over the estimated embryo time.

Gene classification reveals conservation of most transcription factors, and divergence of genes with neuronal function

We next evaluated which types of genes and patterns tend to have conserved or divergent expression (Fig. 3B). Based on manual inspection of numerous genes (Fig. S12), we defined genes as “conserved” in pattern if the gene distance was less than 0.45 or “divergent” for gene distance values above 0.55. Intersecting these categories with the Tau categories, we unsurprisingly find broadly expressed genes were generally highly conserved, while specifically expressed genes were more likely to have divergent expression patterns (Fig. 3B–C). Similar results were found using different, quantile based thresholds (Fig. S13). The specific or patterned genes that were expressed at higher levels were more likely to be conserved: 35% of genes expressed above 80 TPM in either species had divergent expression (JSD_gene > 0.55) while only 14% of genes expressed above 500 TPM in both species had divergent expression. Thus, the relationship between expression breadth, level, and constraint is complex (Fig. S14).

We surveyed what gene functional annotations were associated with gene expression conservation. Essential genes were more likely to have conserved expression patterns: genes with an annotated embryonic or larval lethal phenotype when disrupted by either RNAi or genetic mutation in C. elegans have lower gene distance on average than those that do not have an embryonic phenotype (Fig. 3D). Similarly, the gene distances were inversely correlated with whether the gene is in the same local syntenic position in the two species, is maternally inherited (20), or is phylogenetically older (Fig. S15; (42)). Furthermore, genes annotated with the WormCat terms Chaperone, Muscle Function and Transcription:General Machinery were more conserved; by contrast, genes annotated as Stress Response, Globin, Transmembrane Protein or Neuronal Function were more divergent (41, 43).

To ask whether constraints change across developmental time, we asked which WormCat categories were more likely to have conserved or divergent expression in different breadth (Tau) categories in terminal and progenitor cell types (Fig. 3E). In progenitors, genes annotated with Cell cycle and Development were enriched in the conserved/broadly expressed categories, as expected for rapidly dividing cells. Across terminal and progenitor cell types, certain transcription factor DNA binding domains, especially bHLH and Homeodomain families, were enriched in the conserved categories, consistent with their importance in specification of terminal cell fates (44). Conversely, the Nuclear Hormone Receptor (NHR) family was enriched in the specific-divergent category in progenitors, suggesting that the rapid expansion and divergence of this family in nematodes has been accompanied by expression diversification (45). The chemoreceptor-associated G-protein coupled receptor (GPCR) and Seven Transmembrane (7TM) receptor families were both enriched in the specific-divergent category; like NHRs, these gene families have rapidly expanded in Caenorhabditis (46). Neuropeptides, peptides that often signal through GPCRs, were similarly enriched in the specific-divergent category suggesting possible diversification of this ligand-receptor network.

Genes involved in cilia development and function were enriched for expression divergence in progenitors and conversely for expression conservation in terminal cell types. Investigating expression of cilia genes suggests that this discordance in conservation between early/late cell types is another example of temporal differences between the species. C. elegans expresses most cilia genes slightly earlier than C. briggsae across ciliated neuron lineages (Fig. 3F). As a result, cilia genes are expressed in the same terminal cells in both species, but they are more frequently detected in ciliary neuron progenitors in C. elegans than in C. briggsae, causing them to appear divergent in progenitors. The cilia program in C. elegans is governed by two regulators: DAF-19/RFX and FKH-8/FoxJ (47, 48). We found that daf-19 expression timing is similar in both species, while fkh-8 was delayed in C. briggsae relative to C. elegans.

The germline persistence differences seen for genes like cep-1 and gld-1 and the delay in the cilia program seen in C. briggsae provide two examples of how genes can differ in expression timing across evolution. To assess the extent of this variation, we looked for genes with temporal differences across cell types by adapting a Dynamic Time Warping approach (Fig. S17; see materials and methods;(49)). We defined developmental trajectories for each terminal cell type (e.g. ABa to ABal to ABalp, etc.) and identified genes whose expression differences were decreased when aligned temporally. This analysis identified an average of 98 genes with temporal differences per terminal cell type, with 959 genes identified in at least one trajectory (Table S7–S9; 10.6% of all 9059 genes tested were temporal). About 1/4 of these had similar onsets and showed different persistence in the species while 3/4 had differences in onset timing. Genes with persistence differences were enriched for stress response, proteostasis, and housekeeping functions (Fig. S18). In general, onset timing genes were enriched for terms like Transmembrane protein and Muscle function and were more specifically expressed (Fig. S19). These differences in timing can also explain some of the genes found to have high distances in progenitors, but low distance in terminal cell types (Fig. 2F; Fig. S19). Overall, we find that while most genes show conservation in expression pattern, there are notable differences in expression timing between the species.

Conservation of the transcriptome at the gene regulatory network, but differences in timing

Gene expression changes could be explained by differences in gene regulatory networks. We constructed gene regulation networks using a combination of gene co-expression and transcription factor binding predicted from motifs and existing ChIP-seq data (regulons; see materials and methods; (50–52)). This analysis identified target genes for many transcription factors with known roles in development that were consistent with those roles. For example, the pan-neuronal regulator ceh-48 had targets associated with neuronal function (53), targets of the muscle regulators pat-9 and hlh-1 were associated with muscle function (Fig. 4A; (54, 55)), and the cilia program was predicted to be regulated by both of the cilia regulators daf-19/RFX and fkh-8/FoxJ1 (48). We found high overlap for transcription factor regulons between species both for known and novel regulators with 21 out of the 24 high-confidence transcription factor regulons identified in C. elegans being also found in C. briggsae and 20 out of the 21 high-confidence regulons identified in C. briggsae also found in C. elegans. The high reproducibility of observing these regulons indicates an overall similarity in the trans regulatory environment for these modules (Table S10–S12). However, the regulons identified here represent a small fraction of the total transcription factors, leaving open the possibility that less defined gene regulatory networks may still be divergent between species.

Fig. 4. Inference of gene regulatory networks between species reveals conservation of core regulatory modules.

(A) Enrichment of Worm Category terms for target genes of the transcription factor regulons shows the consistency of the constructed regulons with known gene regulatory networks (The significance of the enrichment is from a Bonferroni corrected Fisher’s exact test (* is p <0.05, ** is p <0.005, *** is p <0.0005). (B) The activity of every high confidence regulon constructed using either motif or CHIP-seq data in C. elegans and C. briggsae. The shown regulon activity is the mean of the single-cell AUC value of the targets of that regulon separated by cell class and embryo time. The values were z-score adjusted and scaled by the maximum value across each regulon. (C) The regulon activity of daf-19 and fkh-8 shown as the mean and 95% confidence interval of the AUC value. (D) The regulon activity of elt-7.

Looking at the activity of these regulons in single cells (calculated as the AUC score; see materials and methods), we found consistent usage of these regulons in the same sets of cell classes across developmental time (R² = 0.947, Fig. 4B). However, there are notable differences in the timing for several of the transcription factor regulons including daf-19 and fkh-8, matching our observations in Fig. 3F, and the intestinal elt-7 regulon, which is delayed in C. briggsae (Fig. 4D). Overall, we find a high level of consistency of the gene regulatory programs, but specific differences in the timing for some regulons.

Comparing expression in homologous cells illustrates the complexity of transcriptome evolution

We also used the single-cell transcription data to examine changes in expression profiles of the individual cell types. Cell types can differ in multiple ways, such as in overall transcript levels, in the expression of cell-type specific “marker” genes, and in the level of expression of non-orthologous genes.

We measured the overall difference in gene expression for each cell type by calculating the Jensen-Shannon Distance on the expression of 1:1 orthologous genes between the two species (referred to subsequently as cell distance; cell type statistics available in Table S13). Cell distance varied between 0.35 and 0.60. As an example, the body wall muscle from the middle of the animal (BWM middle), had relatively high similarity across species with a cell distance of 0.36 (Fig. 5A). Most highly expressed genes were expressed at similar levels in both species including a large group of muscle-specific genes, such as the myosin light chain gene, mlc-3. In contrast, the chemosensory ASG neuron had a cell distance of 0.52 (Fig. 5B), with more variation in the expression levels of highly expressed genes than BWM middle. For example, the homeodomain gene ceh-53 was expressed >1000-fold higher in ASG from C. briggsae compared to that from C. elegans. We tested whether incorporating the non-1:1 orthologous genes changes the ranking of cell divergence by cell distance. To do this, we calculated orthogroup-based expression values by summing the expression of all genes within that orthogroup and recalculated the cell distance. This approach included an additional 3,382 C. elegans genes and 2,953 C. briggsae genes compared to the 1:1 ortholog set. The cell distance calculated on the 1:1 orthologous genes or orthogroups led to a highly similar ranking of cell types (R² = 0.88; Fig. S20).

Fig. 5. Cellular function defines divergence in transcriptomes between homologous cell types.

(A and B) Comparison of the mean TPM values of all orthologous genes between C. elegans and C. briggsae to the log₂ fold change between species showing relatively lower similarity for the ASG neuron and higher similarity for the middle body wall muscle (BWM middle). Gene markers of the respective cell types are overlaid. (C) The cell distance, calculated as the Jensen-Shanon distance (JSD_gene) between all progenitor and terminal cell types. Progenitor cell types are labeled by what cell class they primarily differentiate into. At the transition of the 350 cell to 650+ cell stages, there is a shift in cell similarity that occurs likely as a result of terminal division and differentiation. These patterns are similarly observed in the between species comparisons (Fig. S21–S22). (D) The homologous cell type distances (median and 95% confidence interval of 1,000 bootstrapped values). (E) The cell distance as JSD between the progenitor cell types, organized by division stage shows an hourglass-like pattern of conservation on average.

Calculating cell distance between all cell types between the two species (Fig. 5C) revealed the higher similarity of homologous cell types (diagonals of the heat map) versus non-homologous cell types (off-diagonals), supporting the accuracy of the cell type annotations (see Fig. S21–S22 for within-species comparisons). Cell distance varied both between and within major cell classes (Fig. 5D). The germline, muscle, intestinal and hypodermal cell types were most similar between species, while pharyngeal and mesodermal cells had intermediate cell distance values, and neurons, rectal cells and glia were on average the most divergent. However, cell distance varied substantially within cell classes, especially for the specialized mesodermal cell types, epidermis, pharynx and non-ciliated neurons, potentially reflecting the diversity of functions performed by cell types in these classes. Looking at the progenitor cell types (Fig. 5E), we found that the earliest annotated progenitors have higher cell distances, which then decrease to a minimum mean distance at the 200-cell stage before rising on average in later progenitors and terminal cell types. The higher expression similarity between species at an intermediate time point is reminiscent of the “developmental hourglass” model, with a ‘phylotypic’ time point at the 200-cell stage (56, 57). However, the temporal patterns of cell distance varied between individual lineage fates, reflected in the high variance within each time window. For example, the trajectory leading to the ASG neuron has a classic “hourglass” pattern with a maximum similarity at the 200-cell stage, while transcriptomes of cells in the trajectory leading to head body wall muscles continue to become more similar across time (Fig. S23). This result suggests that progenitor expression conservation across time is complex and may not be easily summarized by a single model (Fig. S24).

Cell type specific expression shows conservation of core functions and specialization of others

While the cell distance compares the overall similarity of the transcriptomic profile of a cell type, it reflects differences in both broadly expressed genes and the cell type-specific genes that determine fate and specialized cellular functions. Thus, we identified and compared cell type-specific ‘marker genes’ of every cell type in both species (Table S14–S15). The number of markers varied between cell types, with germline having the most markers (~2,000), followed by intestine, ciliated neurons and other cell types. Marker counts were well correlated between the species (R² = 0.76; Fig. 6A). Generally, a large proportion of the markers from one species were also a marker of that cell type in the other species. However, cell types varied in the fraction of shared markers; for example 81% of the 456 (1:1) markers of C. elegans mid-body wall muscle were shared by C. briggsae, versus 48% of the 639 C. elegans ASG markers.

Fig. 6. Cell type specific gene expression highlights the complex nature of cell type evolution.

(A) The total number of markers for C. elegans and C. briggsae, colored by cell class and division stage. (B) The fraction of markers that are shared between species showing a strong linear relationship with the cell distance for each terminal cell type. (C) The fraction of “divergent markers” for that cell type that are poorly expressed in that cell type in the other species. (D) The enrichment of Worm Category terms for marker genes in the different conservation categories (The significance of the enrichment is from a Bonferroni corrected Fisher’s exact test (* is p <0.05, ** is p <0.005, *** is p <0.0005). (E) The median dN/dS value of all marker genes of a cell type as a proxy for the degree of protein selection of genes specific to different cell types.

Marker sharing was well correlated with cell distance (Adj. R² of 0.71; Fig. 6B). However certain cell types deviated from this relationship. For example, the somatic gonad precursor cell (Z1/Z4) had lower than expected marker sharing (<45%) given its fairly similar transcriptome overall (cell distance of 0.42; Fig. 6B). A master regulator of C. elegans Z1/Z4 specification, the zinc-finger transcription factor ehn-3 and its distant paralog ztf-16 are redundantly required for somatic gonad development in C. elegans and are classified as markers in C. elegans Z1/Z4 (58). However we see no Z1/Z4 expression of Cbr-ztf-16 and while ehn-3 homology appears complex, none of several candidate C. briggsae ehn-3 homologs were detected in Z1/Z4. This difference suggests a unique role for ehn-3 and ztf-16 in C. elegans somatic gonad specification and raises the question of which, if any of the several C. briggsae specific Z1/Z4 markers encode transcription factors that are C. briggsae-specific regulators in this cell type (Fig. S25).

As the cell type markers that have diverged between the species could indicate a shift in cell-type specific function, we identified “species-specific markers,” defined as markers from one species that are poorly expressed in that cell type in the other species. In general the number of species-specific markers was low (~2–4%) for Germline, Muscle, Intestine and Hypodermis cell types. Conversely the excretory gland, some of the mesodermal cell types, and a few neuron classes stood out as having especially high fractions of species-specific markers (Fig. 6C). Contrasting the function of shared markers versus species-specific markers, we found that shared markers were highly enriched for annotations appropriate for each cell type, such as Muscle Function for BWM middle, or Neuronal Function for ASG (Fig. 6D; Fig. S26–28). In contrast, species-specific markers were instead enriched for terms associated with Environmental Response such as Stress Response (in Intestine) or 7TM receptors (in certain ciliated neurons). These marker sets should provide a useful starting point for functional studies of conserved and species specific gene functions (Table S14–S15).

To assess protein sequence constraint and how it varies across cell types, we compared the relative strengths of selection on protein sequence evolution by computing dN/dS for markers of each cell type (Fig. 6E). The median dN/dS ratio was below 0.1, suggesting that most cell type markers are under purifying selection. But the magnitude of the ratio varied between cell types suggesting variations in strength of purifying selection. The median dN/dS was the highest for germline and early progenitor markers, followed by markers of ciliated neurons and intestine, with an apparently greatest strength of purifying selection for markers of muscle and non-ciliated neurons. Functional constraint of markers as assessed by dN/dS was not well correlated with levels of expression divergence as measured by cell distance. For example the germline had both the highest median dN/dS and the lowest cell distance, while ciliated neurons had a relatively high median dN/dS and cell distance. A pairwise protein substitution metric (K_A) gave similar results (Fig. S29, (59)). This finding emphasizes that genes change independently in expression pattern and protein coding sequence across evolution.

Expression conservation of genes undergoing copy number evolution

Genes undergo duplications and losses, sometimes leading to neo-functionalization or sub-functionalization into new roles (60). We examined the relationship between gene family structure and cell-specific marker genes. While cell type markers were enriched for 1:1 genes, cell types varied substantially in the total number and fraction of their markers with complex orthology relationships in the two species by both developmental stage and cell class (Fig. 7A–B, Fig. S30). Certain terminal cell types, including the amphid sensory neurons had a higher proportion of non-1:1 markers. Many cell types were enriched for categories of non-1:1 marker genes clearly associated with their physiology. For example cytoskeletal genes in muscle cells, such as actin and myosin genes have duplicated independently in C. elegans and C. briggsae, as have metabolism-related genes in intestinal cells (Fig. S28). Some cell types were especially enriched for rapidly evolving gene families; for example the amphid ciliated (sensory) neurons had enrichment for Nuclear Hormone Receptors and the intestine for genes involved in Stress Response. These enrichments likely reflect the specialized functions required for the roles of these cell types in responding to a changing environment, but other non-adaptive explanations are possible (Fig. S28).

Fig. 7. Expression and conservation of genes with non-1:1 orthologous relationships.

(A and B) The number of marker genes in each cell type, shown as the mean count between the two species for all markers (A) and for the non-1:1 markers (B). (C) Enrichment of Worm Category terms for marker genes in the different orthology categories (The significance of the enrichment is from a Bonferroni corrected Fisher’s exact test (* is p <0.05, ** is p <0.005, *** is p <0.0005). (D) Gene expression level and breadth are contrasted for 1:2 gene duplications (where one paralog is a better match to the gene from the other species), with 1:1, many:many, and genes with complex orthology relationships (Uncategorized). Only genes with expression > 80 TPM for at least one copy are shown for the Tau distributions. (E) A maximum-likelihood gene tree of the AP-2 transcription factors compared to their summarized expression reveals novel expression patterns for some duplicates.

Early progenitors prior to ~200 minutes have the largest number and proportion of non-1:1 markers (Fig. 7B). We again observed an hourglass-like pattern in the number of markers, with the lowest number of markers in cells present at ~200 minutes, similar to the time of minimal cell distance seen above. Non-1:1 markers in early progenitors were heavily enriched for genes involved in Ubiquitin-Mediated Proteolysis (F-box genes) and Chromatin Structure, the latter reflecting the duplicated clusters of histone genes (Fig. 7C, Fig. S28). The 1:1 markers at these stages were instead enriched for fundamental developmental processes like mRNA-processing, Development, and Transcription (Fig. 7C). These patterns suggest that different evolutionary constraints act on early progenitors versus late progenitors and terminal cell types.

We examined the expression patterns and levels of cases where we had 1:2 copy relationship between the two species: 452 and 573 duplicated pairs of genes in C. elegans and C. briggsae, respectively (Tables S16–17). On average the 1:2 genes were expressed at lower levels (p<0.0001 by rank-sum test) and more specifically than 1:1 genes (Fig. 7D; p<0.0001 by rank-sum test for 1:2 gene pairs with a minimum expression >80 TPM of all copies in at least one cell type). In most pairs (~70%), one copy was poorly expressed, and when the second copy was expressed (>80 TPM), it was less broad than the better matched ortholog (p<0.0001 by rank-sum test for both level and breadth; for example in sod-2 and sod-3; Fig. S31). This result is consistent with the expectation of gradual loss-of-function of one of the copies due to genetic drift or negative selection against the duplicate that could eventually lead to pseudogenization, although it could also reflect sub- or neo-functionalization. Consistent with the pseudogenization model, 1:2 genes in C. briggsae that have duplicated since their split with the sister species, C. nigoni are enriched for being lowly expressed compared to older duplicates (Chi-squared test p<0.0001; Fig. S32).

For a smaller set of 1:2 genes (~30%), both duplicates were highly expressed and many of these had a highly similar expression pattern. These similarities could be the result of a recent duplication, e.g. C46H11.7/phat-1, but also include cases in which the duplication is older and maintained (e.g., let-418/chd-3). These could be maintained because the higher expression of the duplicate pair is beneficial, or because of additional copy-specific expression in postembryonic stages. In rare cases (~30 orthogroups per species), both paralogs were expressed robustly and in different patterns. For example the AP-2 transcription factors show two examples of species-specific duplications (Fig. 7E); C. elegans aptf-2 has two C. briggsae homologs and aptf-3 and aptf-4 in C. elegans share a single homolog in C. briggsae. In each case one paralog has maintained the (presumably ancestral) expression in the germline and early embryo, while the other has gained cell-specific expression not seen for any other gene in the family.

Finally duplicated genes can change their protein sequence as well as their expression pattern. For the 1:1 orthologs there was only weak correlation (r = 0.28) between gene distance and protein similarity (measured using Smith-Waterman alignment scores), consistent with often independent rates of evolutionary change for protein and expression patterns described above (Fig. S33; (61)). Looking at 1:2 orthologs, we tested whether the more similar ortholog by expression was more similar by protein sequence and found a somewhat higher correlation (r = 0.51 for C. elegans or 0.56 for C. briggsae duplications). Despite this correlation, there were many examples of genes with large expression differences but similar protein sequence, or divergent protein sequence but similar expression (Fig. S33; (62)). These could represent cases of neo- or sub-functionalization but are also consistent with neutral changes in expression due to reduced constraint in the pattern or protein sequence of one or both copies.

Discussion

Despite the evolutionary divergence that has allowed the accumulation of an estimated 1.78 substitutions per neutral site between the species (23), the developmental gene regulation of C. elegans and C. briggsae remains highly conserved. Nonetheless, thousands of individual genes have shifted in expression patterns, with more dramatic changes for genes that are either lowly expressed and/or are highly specific in their expression pattern. We found specific gene families enriched for their divergence in gene expression patterns (neuropeptides, NHRs, F-box proteins, and 7TM/GPCRs; e.g. Fig. 3E). The higher divergence of these gene families largely agrees with previous work that has found cell type specific variation in the expression of the nlp-21 neuropeptide between strains of C. elegans (63) and the frpr-14 neuropeptide between C. elegans and C. briggsae (27). As these are young and expanding gene families (42), our results are consistent with the growing body of evidence that gene duplication drives changes in the gene regulatory program (64). One important caveat is that our two species comparisons were made in relation to within species single cell variation as well as technical replications, rather than by estimates of accumulated neutral phenotypic variation. Given the two species separation (~20 mya; (24)), accumulated neutral mutational variance is likely to be much higher than single cell variation and technical variation; therefore, our inference of evolutionary conservation might be conservative while estimates of putative functional divergence might be liberal. Furthermore, both strains are domesticated isolates (65), and were reared at the same temperature despite having separate preferences (66), potentially influencing the measurement of some transcripts.

Despite thousands of genes showing divergence in their gene expression patterns, genes rarely change in their breadth of expression across the embryo. This may be because the regulatory architecture required to generate broad expression across the animal limits the acquisition of specific expression patterns and vice versa (67). We also found shifts in the temporal expression of many genes across embryonic development. Whether these shifts and expansion of expression breadth represent evolutionary neutral changes or are in response to changing developmental needs will require additional experimentation. Evidence from comparative gene disruption in C. elegans and C. briggsae using RNAi estimated that ~25% of genes have different loss of function phenotypes between these species (68). Combined with our estimate that at least 19% of genes expressed in the embryo have divergent gene expression patterns, this suggests the possibility that many of these gene expression changes have functional consequences.

We leveraged the clear cellular homology between C. elegans and C. briggsae (16, 17) to measure conservation of gene expression between cell types of different classes over developmental time. We established the conserved molecular identity of almost all cell types across the animal. The few exceptions of a cell type found in one species and not the other represent potentially interesting cases that require additional investigation. We found evidence for the hourglass model of development where both the transcriptomic similarity and degree of cell type specific expression appears minimal at the midpoint of embryogenesis (42, 57, 69). However, using the cell type resolution of the dataset, we showed that the maximal point of constraint appears to differ based on terminal cell fate identity, indicating possible tissue specific pressures across development (70). Concordant with the observations that genes involved in neuronal function, such as the neuropeptides and GPCRs are generally more divergent in their gene expression patterns, we find neuronal cell types are also more divergent between species compared to more ancestral cell types such as the intestine and the muscle. Similarly, a new single cell comparison of Caenorhabditis larval neuron transcriptomes finds that neuronally expressed genes, such as neuropeptides, also show species specific expression at later stages (71). These findings indicate that changes in behavior, potentially in response to new environmental requirements, might be a primary driver of cell type evolution for some cell classes.

We showed that cell type-specific transcriptomes can vary independently in multiple dimensions of evolutionary dynamics. For example the germline precursor has a highly conserved overall transcriptome and high sharing of its cell type specific markers, but also a large number of non-1:1 genes and a reduced constraint on cell type marker protein sequence. In contrast, the somatic gonad precursor and amphid neurons express many non-1:1 genes specific to either species. These complexities illustrate the complex functional requirements and constraints on cell type evolution. Further holistic investigation of cell type evolution will likely be required to understand the extant diversity of forms and functions of life.

Overall, these datasets allow for the investigation of specific differences in cell-type specific gene expression across embryonic development in C. elegans and C. briggsae. An explorable version of the datasets is available online (https://cello.shinyapps.io/cel_cbr_embryo_single_cell/).

Material and methods summary

Data collection

New single-cell RNA-seq data was collected for both C. elegans and C. briggsae embryos. The additional C. elegans and the C. briggsae cells were recovered from animals reared, synchronized and collected as described previously in (18). The new data from C. elegans includes a new wild type collection from a later developmental time than (18) along with several collections of embryos where ceh-9, mec-3, or M03D4.4 were mutant. The mutant datasets were filtered to not include any cells where the mutant genes are expressed in a wild type context. The filtered cells from these mutants were further determined to not negatively impact any conclusions from these analyses through extensive testing (see materials and methods).

Orthologous gene comparison

A set of 13,679 genes were classified as being 1:1 orthologs between C. elegans and C. briggsae using a combination of existing definitions from WormBase (72) and retention of synteny between species. Genes with complex orthology relationships (1:2, 1:many, many:many) were determined using OrthoFinder (25, 26).

Initial processing of single-cell data

The single-cell RNA-seq reads were mapped using the 10X genomics Cell Ranger pipeline to a modified version of the WormBase WS290 C. elegans and C. briggsae reference genomes (72). As was previously done for the C. elegans reference genome in (18), the C. briggsae 3’ UTR annotations were similarly extended to improve gene expression quantification. Within all cells, the percentages of reads coming from the background ambient RNA profile was determined and subtracted, similar to (18).

Embryo time estimation

As computed previously in (18), the developmental time of the embryo from which each cell was collected was estimated. Briefly, the log normalized expression data of individual cells were correlated with log normalized bulk-RNA-seq data from (73). For the C. briggsae cells, the estimated embryo time was calculated using the 1:1 orthologous genes between C. elegans and C. briggsae. For both species, the estimated embryo time matched with expected age distributions given the collection scheme (Fig. S2) and with the lineage annotations (Fig. S3)

Cell-type annotation

Cells from both species were integrated and projected into a joint reduced dimensional space using Reciprocal Principal Components Analysis (RPCA) from Seurat V5 (5.1.0; (74–76)). Clusters identified within this joint space using Louvain clustering were matched to tissue classes (such as mesoderm, epidermis, neurons, etc.) using known marker genes (Table S18). The cells were then reintegrated in these subsets using RPCA, and reclustered. Separately, cells were subsetted based on their inferred embryo time, reintegrated using RPCA, and re-clustered for lineage annotation. Using gene markers, trajectory relationships, and the embryo time estimates, the cells were annotated to known lineage and terminal cell type labels (Table S18).

Calculation of summary metrics

The gene expression of single cells were pseudo-bulked based on their lineage and terminal cell type labels into Transcripts per Million (TPM). As the terminal cell types span several hours of development in the embryo, they were split into bins based on their estimated embryo time for subsequent summary metrics calculation. Expression distance and breadth statistics presented here are mean values across these embryo time bins. For both cell types and genes, the distance between the species transcriptomes was calculated using the Jensen-Shannon Distance. Expression breadth across the animal was calculated using the Tau metric from (41).

A full description of the materials and methods used in this work is provided in the supplementary materials.

Data and materials availability:

All raw data is available through the GEO (GSE292756) and SRA through the BioProject number: PRJNA1201888. Previously generated cells from (18) are available through the BioProject number: PRJNA523834. Code used in the generation of the presented figures, tables, and all summarized versions of the dataset are available at the following GitHub link: https://github.com/livinlrg/C.elegans_C.briggsae_Embryo_Single_Cell. An interactive version of the single-cell dataset, available through the VisCello tool, is available at the same link. A Shiny app visualization of the processed data is available at: https://cello.shinyapps.io/cel_cbr_embryo_single_cell/. Databases required to run gene regulatory network inference in C. elegans and C. briggsae are available here at: https://github.com/livinlrg/cistarget_caenorhabditis. All code and data presented in these repositories is additionally stored through Dryad: (77, 78). A user-friendly visualizer of the gene expression data is available at GExplore ((79); https://genome.science.sfu.ca/gexplore).