Comparative Transcriptomics across Nematode Life Cycles Reveal Gene Expression Conservation and Correlated Evolution in Adjacent Developmental Stages

Abstract

Nematodes are highly abundant animals with diverse habitats and lifestyles. Some are free living whereas others parasitize animals or plants, and among the latter, infection abilities change across developmental stages to infect hosts and complete life cycles. To determine the relationship between transcriptome evolution and morphological divergences among nematodes, we compared 48 transcriptomes of different developmental stages across eight nematode species. The transcriptomes were clustered broadly into embryo, larva, and adult stages, with the developmental plastic stages were separated from common larval stages within the larval branch. This suggests that development was the major determining factor after lifestyle changes, such as parasitism, during transcriptome evolution. Such patterns were partly accounted for by tissue-specific genes—such as those in oocytes and the hypodermis—being expressed at different proportions. Although nematodes typically have 3–5 larval stages, the transcriptomes for these stages were found to be highly correlated within each species, suggesting high similarity among larval stages across species. For the Caenorhabditis elegans–Caenorhabditis briggsae and Strongyloides stercoralis–Strongyloides venezuelensis comparisons, we found that ∼50% of genes were expressed at multiple stages, whereas half of their orthologs were also expressed in multiple but different stages. Such frequent changes in expression have resulted in concerted transcriptome evolution across adjacent stages, thus generating species-specific transcriptomes over the course of nematode evolution. Our study provides a first insight into the evolution of nematode transcriptomes beyond embryonic development.

Introduction

Nematodes represent the largest animal phylum on earth and display a vast diversity, with 25,000 species described and ∼10,000,000 estimated (Poinar 2011). Their extensive morphological diversity is a reflection of their trophic resources, lifestyles, reproductive strategies, and living environments. Of the free-living nematodes, Caenorhabditis elegans is the best-studied model organism in molecular and developmental biology. Caenorhabditis briggsae—closely related to C. elegans and with an almost identical morphology (Grün et al. 2014)—is also widely used in comparative studies on nematode evolution and development. Additionally, parasitism is ubiquitous in nematodes and has independently arisen at least 18 times during the group’s evolutionary trajectory (Blaxter et al. 1998; Blaxter and Koutsovoulos 2015; Zarowiecki and Berriman 2015; Weinstein and Kuris 2016). Another nematode genus of particular interests is Strongyloides, the species of which have a unique life cycle in which they alternate between free-living and parasitic generations. Such alterations make Strongyloides a unique and attractive model for studying the evolution of parasitism. A previous study showed that parasitism-associated genes that are expanded and specific to parasitic stages are clustered in specific chromosomal regions, suggesting that they contribute to the regulatory mechanisms of parasite development (Hunt et al. 2016). Some of these parasitism genes expanded across different clades of parasites, indicating convergent evolution at the genomic level (Coghlan et al. 2019); however, the evolutionary relationships among transcriptomes at different stages and parasitic or free-living nematodes remain to be elucidated.

Evolutionary changes occur frequently in organisms through the co-opting of existing traits for new purposes. Which co-opted features change with the emergence of species-specific stages and how they do so at the genetic and regulatory levels are essential questions in evolutionary developmental biology. One theory, the developmental constraint concept, argues that these features limit phenotypic variability and the composition or dynamics of the developmental system (Smith et al. 1985). In nematodes and arthropods, the morphologies and transcriptomes were conserved during midembryogenesis between species within the same phylum (Kalinka et al. 2010; Levin et al. 2012). These observations coalesced into the hourglass model. Because evolution and development are two intertwined processes, constraints and variations in a species’ development may have significant impacts on that species’ evolutionary trajectory. A recent study suggested that some gene expressions changed simultaneously across multiple tissues after speciation, leading to correlated patterns of gene expression evolution and causing the genes to group by species in hierarchical clustering (Liang et al. 2018). Studying constraints in the transcriptomes of stages beyond nematode embryogenesis is of tremendous interest, but remains challenging as developments in each stage can be vastly different across intraspecific generations and interspecific morphologies. One of the first experiments comparing transcriptomes of nematode developmental stages beyond embryogenesis was performed by comparing synchronized transitions from embryo to adult stages in C. elegans and C. briggsae (Grün et al. 2014). It measured fluctuating mRNA and protein expressions across the life stages, and showed that transcript fold changes were conserved during embryo-to-larva transitions. Over the past few years, insights gained from transcriptomic comparisons between developmental stages during and before infection have increased our understanding of parasitism. The recent availability of transcriptomic data from parasitic nematodes (Choi et al. 2011; Stoltzfus et al. 2012; Laing et al. 2013; Hunt et al. 2018; Tanaka et al. 2019) is an exciting resource for identifying the evolution of gene expression throughout development.

The life cycle of nematodes usually consists of one embryonic stage, four to five larval stages, and one adult stage which are separated by molting (Lee 2002; Sommer and Streit 2011). The body size of larvae increases after every molt, eventually reaching sexually mature adult size. Several nematodes have evolved specialized developmental stages, such as a dauer stage whereby the larva undergoes developmental arrest to survive unsuitable conditions, such as a food shortage or high population density. The occurrences of dauer and diapause stages have been associated with gene expression changes in several invertebrates (Flannagan et al. 1998; Bao and Xu 2011; Hand et al. 2016). Unique developmental stages with morphological traits specialized for parasitism also frequently occurred during the evolution of nematode species. These include the sedentary and swollen females in plant-parasitic nematodes and the ensheathed larvae in animal-parasitic nematodes (Lee 2002). In addition, the microfilariae of Brugia malayi, which have a morphology very different from any general life stage of nematodes, migrate to and develop in the mosquito, making B. malayi the intermediate host and transitional insect vector. The transcriptomes of these specialized developmental stages are often distinct from those of previous stages, and these differences in expression mainly come from members of expanded gene families that arose from lineage-specific duplications (Stoltzfus et al. 2012; Baskaran et al. 2015; Hunt et al. 2018).

We hypothesize that, although there is vast phenotypic diversity across the life cycles of different nematodes, these life cycles can be compared, and therefore the levels of conservation between gene expression across developmental stages can be quantified. Hence, high throughput sequencing data across nematodes that previously focused on species-specific differences may be further utilized to reveal the conservation of transcriptomes associated with the life cycles. In this study, we compared the transcriptomes of different developmental stages across several nematode species and profiled the conservation of gene expression in whole worms, particular tissues, and biological processes.

Results

Data Collection and Clustering among Intraspecies Transcriptomes

We collected independently published transcriptome data sets from five to seven developmental stages in eight nematodes: C. elegans and C. briggsae (Grün et al. 2014), Pristionchus pacificus (Baskaran et al. 2015), Strongyloides stercoralis (Stoltzfus et al. 2012), Strongyloides venezuelensis (Hunt et al. 2018), Haemonchus contortus (Laing et al. 2013), B. malayi (Choi et al. 2011), and Bursaphelenchus xylophilus (Tanaka et al. 2019). A summary of the data is shown in figure 1. Grün et al. (2014) profiled the transcriptomes of two Caenorhabditis species during development under the same conditions. The combined data set consisted of 13 stages with one to six biological replicates per stage and 1.5–75 million reads per sample. On average, 82.4% of reads per sample were aligned to corresponding nematode genomes using HISAT2 (Kim et al. 2015) under the same parameters (supplementary table S1, Supplementary Material online).

Fig. 1.

Nematode phylogeny and life cycles. To the left are the phylogenetic relationships among eight nematodes. The abbreviation for each species is shown after the species name, and is used throughout the study. To the right are the life cycles for each species. Different colors correspond to the different nematode lifestyles. The available transcriptome data used in this study are denoted with yellow stripes. Nematode host types are labeled above their life cycles. Notably, stages were defined by molting in all species without proved homology. In total, three free-living nematodes, four animal parasitic nematodes, and one plant parasitic nematode were included in this study. (*¹ L1: First-stage larva, *² L2: Second-stage larva, *³ L3: Third-stage larva, *⁴ L4: Fourth-stage larva, *⁵ LL4: Late fourth-stage larva, *⁶ J1: First-stage juvenile, *⁷ J2: Second-stage juvenile, *⁸ J3: Third-stage juvenile, *⁹ J4: Fourth-stage juvenile, *¹⁰ immature/mature microfilaria, *¹¹ L3 plus: Infectious third-stage larva [isolated from host], *¹² PFemale: Parasitic female, *¹³ PP: Post parasitic, *¹⁴ PFL: Post free living, *¹⁵ L3i: Infectious third-stage larva [isolated from environment], *¹⁶ FLFemale: Free-living female, *¹⁷ D3: Dispersal third-stage juvenile [D3], *¹⁸ D4: Dispersal fourth-stage juvenile).

Rather than recompiling a list of differentially expressed genes between adjacent development stages as done in previous studies (Choi et al. 2011; Stoltzfus et al. 2012; Laing et al. 2013; Grün et al. 2014; Baskaran et al. 2015; Hunt et al. 2018; Tanaka et al. 2019), we were primarily interested in the genome-scale similarities of mRNA expression between development stages. Correlation coefficients between different developmental stages within each species were computed and hierarchically clustered. In general, the transcriptome of the embryo was the most distinct, whereas those of larval stages were similar to those of their adjacent stages (fig. 2). Such clustering patterns remained unchanged with different clustering methods (supplementary fig. S1, Supplementary Material online), downsampling of lowly expressed genes (supplementary fig. S2, Supplementary Material online), gene subsets (supplementary fig. S3, Supplementary Material online), and stages (supplementary fig. S4, Supplementary Material online). In the case of C. elegans, each clustered stage—embryo, adult, and larval—was further clustered into early (L1 and L2) and late stages (L3, L4 and LL4) (fig. 2A). Clustering patterns of larval stages in P. pacificus (fig. 2B) and B. xylophilus (fig. 2C) also supported the early-late partition. This partitioning pattern was not observed for H. contortus (fig. 2D) or C. briggsae (fig. 2E), possibly due to intraspecies variation in the speed of development, which results in worm cultures consisting of individuals with different development stages (Baskaran et al. 2015; Perez et al. 2017). Consistent with previous findings (Baskaran et al. 2015), we also found that specialized phenotypic stages, such as dauer (fig. 2B), were assigned to its adjacent exit stage (larval L4). In addition, the larval stages were divided into two branches: Presence and absence of infectious ability in S. stercoralis (fig. 2F). Similar patterns were found in parasitic larval stages of S. venezuelensis (fig. 2G) and the insect transition stages of B. malayi (fig. 2H). These data suggested that factors associated with parasitism are not as important as factors associated developmental stages in determining similarities of transcriptomes that were analyzed. To compare transcriptomes among stages and different species, orthology between genes was first assigned into a total of 15,835 orthogroups, including 2,548–6,736 single-copy orthologs across 28 species-pair comparisons (supplementary table S2, Supplementary Material online).

Fig. 2.

Transcriptome correlation among nematode developmental stages. Results of hierarchical clustering of Pearson correlations at different stages in single species. Numbers in heatmap indicate Pearson correlation coefficients calculated between pairwise stage comparisons. Colour of the labels demonstrate the lifestyle of corresponding developmental stages. The astericks indicate abrreviated names of developmental stages. The number of genes included in the analysis for each species are as follows: Caenorhabditis elegans 15,156; Caenorhabditis briggsae 16,606; Pristionchus pacificus 19,137; Strongyloides stercoralis 9,823; Strongyloides venezuelensis 12,678; Bursaphelenchus xylophilus 12,259; Brugia malayi 8,899; Haemonchus contortus 14,572.

Interspecies Comparison of Transcriptome across Nematode Development Was Clustered into Embryo, Larval, and Adult Stages

We first computed the correlation between C. elegans–C. briggsae single-copy orthologs across development stages. The development stages were clustered into embryo, larva, and adult stages (fig. 3A). Interestingly, expressions across five larval stages were grouped by species, that is, all C. elegans larval stages were grouped together before clustering with the C. briggsae larval stages. As this particular data set was performed under synchronized conditions (Grün et al. 2014), we included additional published data from modEncode (Celniker et al. 2009), which contains transcriptomes from multiple developmental stages (supplementary table S3, Supplementary Material online). The same clustering was often observed (supplementary fig. S5, Supplementary Material online), even when the cumulative expression of gene families was also compared (supplementary fig. S6, Supplementary Material online), demonstrating that such clustering was robust beyond potential batch effects from different studies.

Fig. 3.

Correlation clustering across nematodes. Hierarchical clustering of Pearson correlations of different stages between (A) two Caenorhabditis species, (B) two Strongyloides species, and (C) Bursaphelenchus xylophilus and Brugia malayi. Numbers in heatmap indicate Pearson correlation coefficients calculated between pairwise stage comparisons. 6,736, 5,394, and 3,109 one to one orthologs were included in the analysis, respectively. Parasitic larva of Strongyloides and the other larva stages were labeled separately.

Similar clustering patterns were also observed in the soil-transmitted gastrointestinal parasitic nematode Strongyloides (fig. 3B). These nematodes are particularly interesting because they alternate between free-living and parasitic generations (Hunt et al. 2016; Hunt et al. 2018). The transcriptomes of S. stercoralis (which parasitizes humans) and S. venezuelensis (which parasitizes rats) were divided into larva and adult stages, but differences were observed in each cluster. Free-living and parasitic female adult stages of each species were grouped together before clustering into a major “adult” group. Transcriptome clusters in the larval branch were instead separated by lifestyles, consistent with the observation that infection ability was already available in their common ancestor (Hunt et al. 2016) and was conserved despite having different hosts. The complete partitioning in larval stage of Caenorhabditis but not Strongyloides was not attributed to sequence divergence, as C. elegans–C. briggsae has higher average protein similarity than S. stercoralis–S. venezuelensis (79% vs. 77%; supplementary fig. S7, Supplementary Material online).

We applied the same analysis to all species pairs to systematically compare gene expressions across nematodes (supplementary fig. S8, Supplementary Material online). Of the 28 possible combinations, clustering of developmental stages in 13 species pairs again revealed three groups corresponding to three broad developmental processes from embryo to adulthood. We found a lower number of ortholog pairs inferred from species of different genera (2,548–4,295) than pairs from the same genus (5,394–6,736). Clustering of transcriptomes between species of different genus partitioned the larval stages further into early and late phases. This was the case for the C. elegans–B. xylophilus and C. elegans–H. contortus comparisons, but not C. elegans–C. briggsae. We were concerned that different pairs of ortholog sets would influence the analysis results, and therefore, repeated the C. elegans and C. briggsae clustering with different one-to-one ortholog lists (C. elegans–B. xylophilus and C. elegans–H. contortus). The clustering results remained identical (supplementary fig. S9, Supplementary Material online, fig. 3A), suggesting greater divergence of transcriptome between C. elegans–C. briggsae larval stages compared with adult or embryonic stages.

Similar environmental pressures lead to recurrent cases of convergence in common genomic adaptations of nematodes of different ancestries (Coghlan et al. 2019). We found one case of transcriptome convergence in the B. malayi–B. xylophilus comparison (fig. 3C). Both species are obligate parasites dispersed by insect vectors. The microfilaria of all stages shared by B. malayi–B. xylophilus followed the three-theme pattern of embryo, adult and larval stages—except adults of B. xylophilus, which were more similar to the late larvae. The late larval stages tended to be grouped by species. In addition, the immature microfilaria and microfilaria congregated with the L2 and D3 larval stages of B. xylophilus. First observed by Patrick Manson in the nineteenth century, the microfilaria of B. malayi is a unique larval form that allows it to move from the human host’s bloodstream to the intestine of the intermediate vector mosquito. Conversely, B. xylophilus is a plant parasitic nematode responsible for pine wilt disease. When faced with an environment high in nematode density and low in food supplies, the B. xylophilus L2 larva molts to form an alternative third stage dispersal juvenile (D3), which is taken up by longhorn beetles and transmitted to another healthy host tree (Tanaka et al. 2019). These distinct larval forms adapted to the insect vector were derived independently in their evolution, suggesting transcriptome convergence.

Orthologs of Cell Type-Specific Genes Exhibit Expression Conservation across Nematode Development

The proportion of different cell types composing the worm body changes dynamically during development, and have been used to determine developmental timings (Perez et al. 2017). The germline cells of C. elegans have ∼4-fold amplification during late L4 stage to young adult stage (Hubbard 2005). Hypodermis performs critical function in both early and late development, including establishing the basic body plan, regulating cell fate specification, guiding cell migration, storing nutrient, and depositing cuticle for molting. To determine if a similar class of orthologs was also overexpressed in specific tissues throughout development, we dissected the entire worm transcriptome to focus on gene expression in the hypodermis and oocytes. We first obtained 938 (supplementary table S4, Supplementary Material online) and 178 (supplementary table S5, Supplementary Material online) genes involved in these two processes in C. elegans based on one previous study of mutants that cause defects in germ cell proliferation (Reinke et al. 2003) and another on cell type-specific RNAseq (Spencer et al. 2011), respectively. We then obtained their orthologs and corresponding expression values in other species. Finally, proportions of the transcriptome from orthologs of C. elegans genes in each stage were found to range from 0.2 to 18% (fig. 4A and B). We identified significant differences in transcriptome proportions across developmental stages (embryo, early larva, late larva, and adult; Wilcoxon ranked sum test; fig. 4C and D). Except for the post parasitic L1 and post free-living L1 stages, expression in the hypodermis was highest during the late larval and embryo stages, followed by the early larval and lowest during the adult stages (fig. 4A). This is consistent with the finding that mutants with defects in genes involved in hypodermis development produced arrested embryos or larvae (Riddle et al. 1997). Orthologs of C. elegans genes that participated in oogenesis had the opposite expression trend, and were expressed the least in the larval stages (fig. 4C). Interestingly, the immature and mature microfilaria are distinct larval forms in B. malayi and also had high expression proportions, which makes sense as this stage comes immediately after the embryo stage and the sheath originates in the envelope of the embryo.

Fig. 4.

Gene expression in specific gene sets. Proportion of gene expression specifically expressed in (A) hypodermis or (B) during oogenesis compared with the rest of transcriptome. The relative expression level on the y-axis was calculated from the proportion of subset gene expressions in all of the one-to-one orthologs’ expression. The stages across nematodes were assigned to five developmental categories: embryo (green), early larva (yellow), late larva (red), adult (blue), and other (gray). Stages in the “other” group were species-specific stages and excluded in the following analysis. The upper-right figures showed the proportional differences among four different developmental categories in (C) hypodermis or (D) during oogenesis. Wilcoxon rank sum test was performed between each category (P value * < 0.05, ** < 0.01, *** < 0.001).

Differential LCE in Nematodes

To further quantify the similarities in the transcriptome across developmental stages, we estimated the level of LCE (Liang et al. 2018) in stage pairs across species. Correlated evolution causes the accumulation of species-specific changes in gene expression after speciation event, resulting in the higher transcriptome similarity between stages within species than the corresponding stage in another species. LCE is a statistical model developed by Liang et al. (2018) that measures the correlation between transcriptomes by estimating the average correlation of phylogenetic contrasts across all genes. High LCE (0.32–0.75) was observed between different larval stages in two Caenorhabditis species. In contrast, LCE was 0.01–0.27 for the embryo stage, which overlaps with the LCE bound of 0.235 retrieved from the simulated stage independence (Liang et al. 2018; fig. 5A). These observations were consistent with the aforementioned finding that different larval stages clustered by species, suggesting that gene expressions in larval stages are correlated. In addition, the finding that LCE was lowest between the embryo and any other stage indicated that the gene expression during embryonic development was distinct from those during larval development and adulthood. In Strongyloides, the LCE values for the six comparisons between larva and adult stages were lower than those for all the other comparisons. LCE values in the comparison between parasitic and free-living females, and the infectious L3 and L3+ (L3 collected from host), were 0.64 and 0.36, respectively (fig. 5B). Consistent with the clustering analysis, LCE results suggested that different lifestyles in adults seemed to have evolved concertedly and not individually.

Fig. 5.

Estimates of levels of correlated evolution (LCE). (A) LCE between Caenorhabditis elegans and Caenorhabditis briggsae in seven developmental stages. (B) LCE between Strongyloides stercoralis and Strongyloides venezuelensis. Free-living L3 (PPL3) and L1 (PFLL1) in S. stercoralis were excluded. The PPL1 stage in S. stercoralis was assigned as the L1 stage in the comparison because it has the same post parasitic features as the L1 stage in S. venezuelensis. (C) LCE among all the species. The parasitic and free-living generations in Strongyloides were separated to compare them to developmental stages in other species. Stage comparisons with fewer than three species pairs were excluded. The lowest LCE value with theoretical P-value < 0.05 was 0.235. Colors indicate the different categories of stage comparison. All larval stages were classified into larva categories to test the level of co-evolution.

Next, we sought to determine whether concerted evolution occurred across nematodes of different genera (fig. 5C). Species-specific developmental stages, such as microfilaria in B. malayi and D3 in B. xylophilus, were excluded because they lacked equivalent stages in the other species. Four of the five comparisons between larval stages showed high LCEs (0.27–0.53) whereas six of the eight comparisons between larva and embryo or adult had low LCEs (–0.15 to –0.11). Co-evolution of larval stages within a species seemed to be a general phenomenon in nematodes. Of the larval stages, L3 had on average the lowest LCE, whereas the other stages have average LCEs of 0.38–0.51. This suggested that L3 was the most individualized stage. In parasitic nematodes, infectious ability was frequently observed in the third larval stage. This preference for gaining parasitic capabilities provoked the question of whether L3 in parasitic nematodes has evolved independently since early evolutionary history. Our findings on the LCE of parasites suggested that the answer is no, as L3 was not individualized and instead evolved in concert with the adjacent stages after L3 (supplementary fig. S10, Supplementary Material online).

Orthologs Are Frequently Expressed in Different Developmental Stages

To further quantify the differences in transcriptomes among species, we calculated the Z-score angle-sorted value index for temporal sorting (ZAVIT; Levin et al. 2016), which ranked each gene’s relative expression profile across developmental stages to the rest of transcriptome. Next, each gene in each species was categorized based on its ranking and differential expression analysis across developmental stages (fig. 6A). Each ZAVIT category was then described based on its features (supplementary table S6, Supplementary Material online). For example, 651 C. elegans genes were considered expressed in larval as they had on average 2 times more expression (transcript per million; 41.2 in larval) than the other two broad developmental stages (embryo 27.4 and adult 11.4; supplementary table S6, Supplementary Material online). Within this category, 77 and 82% of genes were differentially upregulated in larval stages when compared with embryo and adult stages, respectively (supplementary table S6, Supplementary Material online). The expression profiles for C. elegans and C. briggsae were similar, both showing a stage-like pattern in the embryo, all larval, and adult stages (supplementary fig. S11A, Supplementary Material online). The majority of genes were expressed in multiple stages (72.5% and 63.4% in C. elegans and C. briggsae, respectively). To examine whether there is an evolutionary preference for any particular category of genes, we assessed the differences in expression categories between species. Interestingly, only just half (50.03%) of C. elegans orthologs were expressed in the same category in C. briggsae (fig. 6A); the proportion of genes that shifted across developmental profiles varied from 20.6% (in embryo + larva) to 86.1% (in larval + adult). Interestingly, genes of the ortholog pairs expressed in the latter category in both Caenorhabditis species exhibited ratios of the nonsynonymous substitution rate to the synonymous substitution rate (d_N/d_S) significantly lower than four other expression profiles (fig. 6C), implying stronger purifying selection on coding sequences of genes that maintained the same role, despite being in the category with the highest switching. Enrichment of Gene Ontology revealed the significant terms including the “small molecule metabolic process” and “purine nucleoside monophosphate metabolism” (supplementary table S7, Supplementary Material online). Small-molecular signaling has been extensively studied in C. elegans for its important roles across multiple aspects of development and behavior (Ludewig 2013), whereas purine homeostasis was recently revealed to be necessary for developmental timing in C. elegans (Marsac et al. 2019). Both conservation of expression category and higher purifying selection of these genes further imply their functional importance across Caenorhabditis genera.

Fig. 6.

Expression profiles and sequence divergences during development in species of Caenorhabditis and Strongyloides. Expression profiles of (A) Caenorhabditis elegans and Caenorhabditis briggsae; and (B) Strongyloides stercoralis and Strongyloides venezuelensis. Expression profiles were categorized into five developmental categories in each species. Sequence divergence of different expression categories in (C) C. elegans and C. briggsae; (D) S. stercoralis and S. venezuelensis. The five boxes underneath include the genes with the same expression profiles in both species. Wilcoxon test was performed to test the difference in d_N/d_S between categories. Asterisks were used to represent the P-value (* < 0.05, ** < 0.01, *** < 0.001). Comparisons with P-value > 0.05 were not labeled.

Expression patterns in free-living and parasitic life cycles were complex and included multiple possible combinations, so we first empirically assigned 88.0% and 91.4% of genes in S. stercoralis and S. venezuelensis, respectively, into four categories based on when they were expressed (fig. 6B, supplementary fig. S11B, Supplementary Material online): In free-living generations, throughout adulthood, parasitic larval stage only, and throughout larval stage (supplementary table S6, Supplementary Material online). Interestingly, the rest of genes were assigned to expression groups in parasitic larval + free-living adult or parasitic larval + parasitic adult in S. venezuelensis and S. stercoralis, respectively. Strikingly, the majority of genes (87.0% in S. stercoralis and 82.4% in S. venezuelensis) were expressed in more than one developmental phase, whereas over half (54.8%) of Strongyloides orthologs were assigned to different developmental categories. A higher proportion of Strongyloides’ genes were expressed at multiple stages than Caenorhabditis, but showed similar levels of developmental switching. The d_N/d_S ratios were calculated for each category, and none exhibited a significantly higher ratio than genes that exhibited a different expression category (fig. 6D).

Discussion

Very little is known about transcriptome conservation between nematodes beyond the embryonic stage. In this study, we compared the developmental transcriptomes of eight nematode species with similar developmental stages. These species have many morphological and developmental differences and a diversity of living environments, lifespans, and host types. The transcriptomes used in this study came from data sets from multiple sources, but all clustered into three broad stages (embryonic, larval, and adult) across nematodes’ entire life cycles. One major concern was that batch effects would lead to systematic differences between data sets (Fei et al. 2018) that could not be separated from species effects (Leek et al. 2010). Attempts to treat batch effects would also remove biological signals from species. Despite this caveat, the same pattern was observed when we incorporated transcriptomes of the same stage from multiple sources of C. elegans, suggesting that the biological signals were robust enough to make up for the technical variations across different studies. Another challenge presented in our study was that the development stages were incomplete in some nematodes, which led to reduced resolution in our analyses. Imperfect synchronization of worm cultures was observed during the staging protocol in P. pacificus, and such issue may be applicable to all nonmodel species. In addition, the definition of developmental stages beyond embryogenesis to adulthood in nematodes was only loosely defined by molting. Even in the model Caenorhabditis, we still have concerns related to perfect synchronization of developmental stages. For instance, the culturing environment may have been more stressful for one species than the other. Further experimental work across all nematode genera is needed to characterize patterns of conservation across their entire life cycles.

Cell-specific information is critical for deciphering how molecular mechanisms control the phenotypes of multicellular animals. Large scale research on gene expression in mammalian organ development suggests that organs become increasingly more distinct and the breadth of gene expression gradually decreases during development (Cardoso-Moreira et al. 2019). This evidence supports the theory of von Baer that morphological differences between species increase as development advances (Abzhanov 2013). So far, gene expression has been determined in specific cell types of C. elegans by dissecting different worm tissues (Spencer et al. 2011) and machine learning-based predictions (Kaletsky et al. 2018). Across nematodes, we revealed that orthologs of oogenesis and hypodermis-specific genes in C. elegans also displayed a shared pattern of transcriptome proportions across developmental stages. Identifying tissue-specific genes in other nematodes using data from C. elegans alone will inevitably underestimate the proportions of tissue expression contributing to whole-worm transcriptomes. We attempted to minimize such bias by normalizing total gene expression to only one-to-one orthologs. Although one-to-one orthology may capitulate a subset of developmentally conserved genes, we have shown that they provided initial insights into gene expressions during tissue development across nematodes.

We examined the stage clustering further by inferring the level of LCE, a statistical method originally intended to quantify correlation between tissue transcriptomes (Liang et al. 2018). Although the LCE estimation revealed strong concerted evolution between individual larval stages, an alternative explanation may be the imperfect synchronization of worm culture in nonmodel organisms. For instance, the staging protocol in P. pacificus was observed in major developmental transcriptome clustering with a mixture of early larvae, late larvae, and adult (Baskaran et al. 2015). Nevertheless, this may not be the case, at least in the Caenorhabditis data set, as the largest proportion of expression profiles were nonadjacent developmental stages (fig. 6A). At least half of the genes in nematodes were expressed at multiple stages, and half of their orthologs were found to be expressed at multiple other stages; this suggests that a change in a gene’s expression during development may rapidly lead to transcriptome divergence after speciation.

Using Strongyloides, a unique model to investigate parasitism based on the fascinating features of both free-living and parasitic generations, allowed us to systematically examine the differences between the same developmental stages in different life style strategies. We found that the transcriptomes of Strongyloides can be categorized based on developmental stage instead of lifestyle, which is in contrast to the observation that up to 20% of the genes are differentially expressed between parasitic and free-living females (Hunt et al. 2018). The reason behind the two different observations is that that the majority of these differentially expressed genes were duplicated in the Strongyloides lineage (Hunt et al. 2018), and we focused on one-to-one orthologs. We have to a certain extent identified a theme of developmental conservation across nematodes, and shown that the specialization into parasitic stages was the result of duplication events in gene families, as is evident in many nematode genomes (Hunt et al. 2016; Coghlan et al. 2019). We speculate that altering the expression of a gene to adapt to a new environmental niche may take place before genomic innovation without reducing much fitness.

The third larval stage is thought to be a common developmental point for obtaining infectious ability (Lee 2002). Gene expression in third larval stages was clustered in parasites, but not with all corresponding stages, especially the free-living L3 and diapause ones. This suggests that the similarities in transcriptomes among parasitic stages were not inherited from their common ancestors but through convergent evolution of having similar selection pressures to tolerate the host environment. The results of the expression divergence analysis show that genes tend to be expressed multiple times over the course of the developmental process. We propose that the life cycle of the nematode common ancestor consisted of an embryo stage and an adult stage, with several larval stages in between. The specialized larval stages—such as the dauer, filarial, and sheathed larva stages—may have independently evolved in response to biological requirements over evolutionary time.

Our study adds to recent efforts to sequence and compare genomes across many nematodes (Coghlan et al. 2019) by providing a first step toward revealing life cycle conservation and convergence at the transcriptome level. The most striking pattern was perhaps the finding that some patterns are conserved in species that diverged many millions of years ago and have drastically different lifestyles. Our results also provide initial insights into how ancestral life strategies, such as parasitism, evolved to become specialized. Future large-scale synchronized experiments across life cycles as well as tissue-specific or single-cell transcriptomes between nematodes may further elucidate life cycle evolution in nematodes.

Materials and Methods

RNA-Seq Mapping and Normalization

A description of locations where RNA sequencing (RNA-seq) reads were downloaded is presented in supplementary table S1, Supplementary Material online. RNA-seq reads were first trimmed using Trimmomatic (v0.36; parameter: LEADING:5 TRAILING:5 SLIDINGWINDOW:3:15 MINLEN:34; Bolger et al. 2014) to remove the adaptor and leading, tailing, and low quality sequences. Trimmed reads from each species were mapped to a corresponding genome assembly downloaded from Wormbase (ver. WS269; Lee et al., 2018) using HISAT2 (ver. 2.2.1; Kim et al. 2015). Raw gene counts were assigned using featureCounts (v.1.6.3; Liao et al. 2014). The raw counts of orthologous genes in all samples were transformed into TPM (transcripts per million), and the median of replicates were calculated to represent the raw gene expressions of developmental stages in each species. To normalize the data, we initially removed the 25% lowest-expressed genes in each species using the sum of samples. Next, we performed the “withinLaneNormalization” function in EDASeq (v2.18.0) (Risso et al. 2011) to remove GC bias for each gene, and transformed the expressions by log2. Considering that our data were collected from multiple studies, we accounted for study design batch effects using the “ComBat” function from the sva (v3.32.1) (Leek et al. 2016). In the case of species-paired comparisons, both orthologs below the 25% expression category were removed for further analyses. Pearson correlation coefficient of normalized transcriptomes in different developmental stages within and between species was determined using the “corr” function in R (v3.6.0; R Core Development Team 2019). The heat maps of correlation matrices were hierarchically clustered with the average agglomeration method.

Phylogenetic and Evolutionary Analysis

Orthology of proteomes from species investigated in this study was inferred using OrthoFinder (v2.2.7; Emms and Kelly 2015). If multiple isoforms exist for a given gene, only the longest or major isoform was chosen for analyses. A maximum likelihood phylogeny was constructed by the concatenated amino acid alignments of 2,205 single-copy orthologs across eight nematodes using RAxML (v8.2.11; -m PROTGAMMAILGF -f -a; Stamatakis 2014) with 500 bootstrap replicates. To calculate sequence-based metrics, sequences of single-copy orthologs were retrieved and aligned using TranslatorX (version 1.1; Abascal et al. 2010). We identified the synonymous (dS) and nonsynonymous (dN) substitution rates using Codeml in PAML (v4.9; parameter: runmode = –2, seqtype = 1, CodonFreq = 3, fix_omega = 0; Yang 2007). The 209 oogenesis-enriched genes were defined by Reinke et al. (2003). The 172 hypodermis-specific genes in C. elegans were defined by Spencer et al. (2011). One-to-one orthologs of these C. elegans genes in other species were retrieved. For each species pair comparison, transcriptome per million (TPM) were calculated using only the one-to-one orthologs where total transcriptome counts equal to the sum of read counts in these genes.

Comparative Transcriptomic Analysis

Levels of correlated evolution (LCE) between transcriptome data sets was calculated according to Liang et al. (2018; https://github.com/cloverliang/LCE). We applied the ZAVIT method (Levin et al. 2016) to organize gene expressions across the developmental process. ZAVIT sorted the standardized gene expressions by the relative order of the first two principal components. The standardized expression profiles were obtained by subtracting the mean and dividing by the standard deviation of the orthologous gene expression for all developmental stages. Principal component analysis was performed to standardize gene expression profiles into ordered genes in a circle, and the angle computed by an invert tangent from the origin represented the temporal expression profiles during development. The first gene in the sorted standardized gene expression profile plot was defined according to the sliding window below.

All developmental stages were first categorized into embryo, larva, and adult stages. Additionally, free-living or parasitic stages were considered separated in the Strongyloides comparison. This yielded three and four stages in the Caenorhabditis and Strongyloides comparisons, respectively. Differential expression analyses between developmental stages were performed using DESeq2 (v1.24.0, P_adj < 0.05; Love et al. 2014). 81.3–82.2% and 63.1–70.2% of genes were first unambiguously sorted into seven and nine expression categories in the Caenorhabditis and Strongyloides comparisons, respectively, based solely on DESeq2 results. These results were used to further split the ZAVIT-sorted genes into five major expression categories. The boundaries were set by choosing the 5% and 10% quantile of the DEseq2 categories in the Caenorhabditis and Strongyloides comparisons, respectively. We were interested in the relative order of each gene’s position within the transcriptome, and this approach allowed for the placement of certain genes—for example, the housekeeping genes, which were expressed throughout development, were sorted relative to the rest of the transcriptome. The starting position of the parasitic larva + parasitic adult expression category in S. stercoralis and S. venezuelensis were determined solely by the 95% quantile of the adjacent free-living adult + parasitic adult DEseq category, as genes positioned around this region tend to exhibit no significantly different expressions in DEseq2 analyses. Gene Ontology enrichment was performed using topGO (v2.36.0; Alexa and Rahnenfuhrer 2019) with GO annotations downloaded from WormBase.

Acknowledgments

I.J.T was supported by Career Development Award AS-CDA-107-L01, Academia Sinica.

Author Contributions

I.J.T. conceived the study. M.R.L. carried out the majority of analysis with help from C.K.-L. and B.-Y.L. M.R.L., and I.J.T. wrote the manuscript.