Main

Multicellularity has evolved multiple times in eukaryotes13. This evolutionary transition often resulted in relatively simple life forms, but in some lineages this transition was followed by a series of evolutionary innovations, resulting in more ‘complex multicellular’ organisms with distinct cell and tissue types, intercellular communication and an intricate developmental programme1,2. The emergence of complex multicellularity is thus thought to be a rare event, having occurred independently in animals, fungi, plants, red algae, and brown algae1. With the rise of complex multicellular lineages, a major question is how developmental processes accommodate evolutionary change.

A recurring pattern of evolutionary conservation and variation across developmental stages during the life cycle of multicellular organisms was already observed by nineteenth century comparative embryologists, who noticed the marked morphological similarity between embryos14,15,16,17. This observation was more recently revisited at the molecular level, where the morphological pattern of evolutionary conservation and variation is supported by an analogous pattern at the transcriptomic level18,19,20. Two models have been proposed to describe how conserved developmental processes accommodate evolutionary change: the early conservation model and the developmental hourglass model (as well as hybrids of the two models). According to the early conservation model, evolutionary change is increasingly permitted as embryogenesis proceeds, which presents a low–mid–high pattern of evolutionary novelty and originates from von Baer’s third law of embryogenesis14. By contrast, the developmental hourglass model proposes that evolutionary change is restricted in the mid-phase of embryogenesis, presenting a high–low–high pattern of evolutionary novelty. This model is motivated by morphological differences observed in the early phases of embryogenesis (for example, diversity in embryo cleavage patterns), similarity in mid-embryogenesis (as embryos converge on a basic body plan) and differences in the later phases (as embryos acquire further species-specific features)3,4.

Using transcriptome novelty as a quantitative readout for evolutionary novelty, some early studies have reported early conservation patterns21,22,23, whereas more recent studies have reported hourglass patterns across various multicellular eukaryotic lineages using bulk5,6,7,8,9,10 and single-cell transcriptomics24,25,26,27. It should be noted that different biological properties such as pleiotropically expressed genes, mutational robustness, inter-embryo expression variability, chromatin accessibility and enhancer conservation may follow different models28. The empirical findings at the transcriptome level further tie in with theoretical modelling, which supports the narrative of the natural emergence of hourglass-like structures in complex evolving systems29,30,31,32.

However, our understanding of the generality of the developmental hourglass across complex multicellular eukaryotes is incomplete without considering the brown algae. Brown algae (also called brown seaweeds) belong to the stramenopiles, a large supergroup of organisms that are only distantly related to animals, land plants and fungi33. Notably, brown algae independently evolved complex multicellularity 450 million years ago11,12,34, and have since become the third most morphologically complex lineage on earth, comparable to plants in terms of size, number of cell and tissue types and developmental complexity33. In addition, brown algal species vary in relative morphological complexity. For example, the ‘morphologically simple’ filamentous Ectocarpus is composed of up to eight cell types35, and is capable of multicellular growth and differentiation without canonical embryogenesis36. By contrast, the ‘morphologically complex’ kelps (in the broad sense, including Laminariales and members of Tilopteridales) and Fucales undergo obligatory, canonical embryogenesis to generate metres-long adult individuals composed of dozens of cell types37. By harnessing the diversity of morphological complexity in brown algae, we can disentangle the effect of multicellular development per se versus embryogenesis on transcriptome evolution patterns. Furthermore, sexual systems also evolved independently in brown algae38. During the course of sexual development in brown algae, most species (for example, Ectocarpus) alternate between haploid (gametophyte) and diploid (sporophyte) generations, each consisting of morphologically distinct, multicellular forms connected by three different unicellular stages: gametes, meiospores and mitospores35 (Fig. 1). We can thus distinguish the potential role of selection in gametes (for example, due to sperm competition) from unicellularity (that is, bottlenecks during the life cycle)39. Alongside the convergent evolution of complex multicellularity, these lineage-specific features make brown algae a unique and powerful system to distinguish overlapping processes seen in animal and plant development.

Fig. 1: Developmental and morphological diversity in brown algae.
figure 1

a, Phylogenetic position of F. serratus, F. distichus and Ectocarpus in a simplified eukaryotic tree of life. An arrow marks the independent origin of complex multicellularity in brown algae. The asterisks mark other lineages that evolved complex multicellularity. b, The life cycle of Fucus species. matSP denotes the sexually mature sporophyte (2n). F, female; m, male. c, The life cycle of Ectocarpus. Unicellular stages are highlighted in grey. GA denotes multicellular gametophytes (n) and (P)SP denotes both multicellular sporophytes (2n) and morphologically identical partheno-sporophytes (n).

Full size image

Leaning on this unique natural history, we explore brown algal species exhibiting distinct levels of morphological complexity, to investigate the existence of a developmental hourglass pattern in this group of complex multicellular eukaryotes. If a molecular hourglass pattern does shape development in brown algae, a question arises as to whether the same hourglass model underlies the development of all complex multicellular life.

Here, we propose an experimental design to test whether a molecular hourglass pattern is underlying brown algal development. To approach this, we quantified gene expression levels across key ontogenetic stages for three distinct brown algal species with external development: Fucus serratus, Fucus distichus and Ectocarpus sp. We selected these species to cover the broad diversity of morphological complexity in this group of eukaryotes and because they develop without contamination from parental tissues (Fig. 1a). The Fucales are morphologically complex seaweeds with a well-described embryogenesis that occurs highly synchronously40. F. serratus has separate male and female sexes, whereas F. distichus is a co-sexual species—that is, the same individual produces male and female reproductive structures (Fig. 1b). As a comparative model, we used the filamentous brown alga Ectocarpus, which alternates between two simple but morphologically distinct and independently developing forms, the gametophyte and sporophyte. Ectocarpus also presents a range of uni- and multicellular stages but not necessarily a canonical embryogenesis35 (Fig. 1c). To further broaden the species diversity, we analysed transcriptomic data from a subset of developmental stages in two different kelp species, Laminaria digitata and Saccorhiza polyschides. Both species alternate between morphologically simple gametophytes (which mirrors Ectocarpus development) and morphologically complex sporophytes (which mirrors Fucus development including embryogenesis). 

Transcriptome evolution in Fucus embryogenesis

We used an evolutionary transcriptomics approach41 to test the developmental hourglass hypothesis for the embryogenesis of the two Fucus species. We first collected stage-specific RNA-sequencing (RNA-seq) data (Supplementary Table 1) and assigned phylogenetic ages to each protein-coding gene using GenEra42 (Extended Data Fig. 1). Combining the expression and evolutionary information, we computed the transcriptome age indices (TAIs) for each stage, which quantifies the weighted mean of the gene age with its transcript expression level5,41. In total, we captured the evolutionary and expression data for 8,291 genes in F. serratus and 7,907 genes in F. distichus (Methods and Supplementary Table 2).

TAI profiles across embryogenesis revealed a transcriptomic hourglass pattern in both Fucus species (Fig. 2a,b). The TAI profiles were robust to all RNA-seq data transformations, consistently returning significant P values (<0.05) for both the flat line and the reductive hourglass tests43 (Extended Data Fig. 2 and Methods). We further tested the robustness of the observed hourglass patterns by removing genes with ‘noisy’ expression profiles using noisyR44 and confirmed that the resulting hourglass patterns in both species remained largely significant (Extended Data Fig. 3). We note a shift in the timing of the developmental stages between the two species, where the transcriptome of the F. distichus stage with the lowest TAI (S4.5) was most similar to the F. serratus stage with the lowest TAI (S4) (Extended Data Fig. 4). Extended Data Fig. 4 also highlights transcriptome divergence in early and late stages. Together, these results provide strong evidence that analogous to animals and plants, a developmental hourglass pattern is also shaping embryogenesis of both Fucus species.

Fig. 2: Presence of molecular hourglass patterns in Fucus embryogenesis.
figure 2

a,b, TAI profile across embryo stages in F. serratus (a) and F. distichus (b). A lower TAI marks an older transcriptome (that is, composed of evolutionarily older genes) and vice versa. c,d, The mean expression profile of each phylostratum (PS) across embryo stages in F. serratus (c) and F. distichus (d). Note that expression is restricted in the category of evolutionarily youngest genes (PS 8) in both species, as indicated by red arrows. TPM, transcripts per million. e,f, TSI profile of embryogenesis in F. serratus (e) and F. distichus (f). A lower TSI marks a transcriptome composed of more broadly expressed (temporally pleiotropic) genes and vice versa. 50,000 bootstraps were used to compute the s.d. in a,b,e,f and the 95% confidence interval c,d.

Full size image

Of note, we observed that the expression of evolutionarily young genes (that is, genes associated with the origin of complex multicellularity in brown algae in phylostratum (PS) 7, and species-specific genes in PS 8) were markedly lower during the waist of the hourglass, as opposed to the expression of older genes being higher at this stage (Fig. 2c,d). Therefore, the repression of expression of young genes may underlie the waist of the hourglass, recapitulating observations in other systems5,9,45.

It has been proposed that the conserved transcriptome composition in the waist of the hourglass is caused by higher pleiotropy of genes expressed in these stages46,47. We examined stage-specific expression using tau as an estimate for temporal pleiotropy48,49, and computed the resulting transcriptome specificity index (TSI) profile across developmental stages. Stages corresponding to the waist of the hourglass (S4 and S4.5 for F. serratus and F. distichus, respectively) were represented by more broadly expressed genes (that is, low tau) whereas early and later developmental stages were characterized by more stage-specific genes (Fig. 2e,f and Supplementary Table 3). Notably, gene expression breadth correlates with the number of protein–protein interactants, developmental essentiality and expression quantitative trait locus-based pleiotropy measures in other systems46,50. In sum, more broadly expressed (potentially pleiotropic) genes are associated with the evolutionarily conserved stages of development in both Fucus species, consistent with findings in animals46,47.

Finally, we investigated the possible biological processes that underlie transcriptome evolution patterns during Fucus embryogenesis. Of note, the waist of the hourglass in both Fucus species paralleled a major ontogenetic transition, from a ‘cell-type differentiation’ stage, in which the algal body plan is established, to a more ‘proliferative’ stage, in which development is mainly characterized by somatic cell divisions leading to expansion in the size of the organism (Extended Data Fig. 5). This finding mirrors similar observations in animals and plants, in which transcriptomic hourglass patterns mark major ontogenetic transitions (for example, cell fate acquisition to differentiated cell growth)18,45.

Together, our observations demonstrate that both Fucus species display a transcriptomic hourglass pattern. The conserved mid-embryonic period is characterized by reduced expression of evolutionarily young genes and a relatively higher expression of more broadly expressed genes, and corresponds with a major developmental transition from cell fate determination to cell proliferation.

Transcriptome evolution in Fucus adults

In animals and plants, the degree of transcriptome conservation also differs in life cycle stages outside embryogenesis (such as between tissues or sexes). For example, evolutionarily young genes are disproportionately expressed in the testis of nematodes, flies and mammals51,52,53,54, and in male reproductive cells in plants55,56. To test whether these patterns are present in brown algae, we examined the TAI across different adult tissues of the two Fucus species, as well as between sexes (Fig. 3a).

Fig. 3: Transcriptome evolution landscape across sex and tissues in adult Fucus.
figure 3

a, Outline of adult (mature sporophyte) tissues and gametes in Fucus species. Note, F. serratus has separate male and female sexes, whereas in the co-sexual F. distichus, the same individual produces both male and female gametes. Reptip, reproductive tip; vegtip, vegetative tip. b, TAI profile across adult tissues and gametes in F. serratus. c, TAI profile across adult tissues and gametes in F. distichus. A lower TAI marks a transcriptome composed of evolutionarily older genes and vice versa. We used 50,000 bootstraps to compute the s.d.

Full size image

In F. serratus males, TAI values differed significantly between tissues (flat line test; P = 8.57 × 10−4), with reproductive tissues (reproductive tip) exhibiting a markedly higher TAI compared to vegetative tissues (vegetative tip, holdfast and stipe) (pairwise TAI test; P = 0.00534) (Fig. 3b). This difference was also observed in F. distichus (pairwise TAI test; P = 0.00240) (Fig. 3c). By contrast, in F. serratus females, TAI values did not differ significantly between vegetative and reproductive tissues (pairwise TAI test; P = 0.407) (Fig. 3b). We also noticed that sexual differences of TAI were more pronounced at the reproductive tip in F. serratus adults, with males having a markedly younger transcriptome compared to females (pairwise TAI test; P = 2.59 × 10−6). Notably, the stipe (in both species) also presented relatively young transcriptomes, specifically in males of F. serratus (Supplementary Table 4). Therefore, similar to animals and plants, the levels of conservation of transcriptomic patterns are variable across the life cycle of Fucus, with Fucus males (particularly reproductive tissue) displaying an evolutionarily younger transcriptome.

Transcriptome evolution in Ectocarpus

In contrast to Fucus, Ectocarpus morphology is substantially simpler, with fewer cell and tissue types, developing in absence of canonical embryogenesis35,36. Its life cycle comprises two morphologically distinct, free living multicellular generations, the gametophyte and the sporophyte (see Fig. 1c), each composed of three to five cell types35. During this alternation of generations, a total of three types of unicellular units are produced: gametes, meiospores and mitospores (see Fig. 1c), which enables us to disentangle the effect of being a ‘gamete’ per se from the effect of being a unicellular unit. Furthermore, Ectocarpus gametes can develop parthenogenically (without gamete fusion) into haploid adults (parthenosporophyte) whose morphology closely resembles that of a diploid sporophyte36 (see Fig. 1). Thus, Ectocarpus is a powerful comparative system to test whether the hourglass pattern seen during Fucus embryogenesis is the result of constraints imposed by multicellular development per se or whether this hourglass pattern of transcriptome conservation is tied to a specific embryogenesis process present in Fucus but not in Ectocarpus.

We examined the developmental transcriptome during the Ectocarpus parthenogenetic life cycle by profiling the transcriptomes of the three unicellular stages (gametes, meiospores and mitospores), three stages during gametophyte development (immature, mature and senescent), and three stages in parthenosporophyte development (early, immature and mature), for both male and female lines (Extended Data Fig. 5 and Supplementary Table 1). As done in Fucus, we performed gene age inference (Extended Data Fig. 1 and Methods), and computed the TAI values from 11,571 genes at each developmental stage (Supplementary Table 2). This analysis revealed that Ectocarpus unicellular stages in both males and females exhibited a significantly higher TAI compared with the multicellular stages (Fig. 4a and Extended Data Fig. 6), and this pattern was consistent across RNA-seq data transformations (Extended Data Fig. 7). When restricting the reductive hourglass test to multicellular gametophytic or partheno-sporophytic development, only the male partheno-sporophytes returned a significant hourglass shape (reductive hourglass test; P = 4.22 × 10−15). Since this pattern is not consistent in other instances of multicellular development (male gametophyte, female gametophyte and parthenosporophyte) (Supplementary Table 4), we interpret that multicellular stages in these filamentous brown algae have a lower TAI overall, rather than displaying a canonical transcriptomic hourglass. Moreover, conservation in the multicellular stages according to the TAI patterns was recapitulated when computing transcriptome distance/similarity between life cycle stages in Ectocarpus (both males and females) and the embryo stages in Fucus (Extended Data Fig. 8).

Fig. 4: Transcriptome evolution dynamics across the life cycle of Ectocarpus.
figure 4

a, TAI profile across Ectocarpus life cycle stages. A lower TAI marks a transcriptome composed of evolutionarily older genes and vice versa. b, TDI profile across Ectocarpus life cycle stages. Note that the TDI captures the degree of purifying selection because it is based on the dN/dS ratio of one-to-one orthologous genes between Ectocarpus sp. 7 and Ectocarpus subulatus, >99% of which is less than 1. Thus, a lower TDI marks a transcriptome composed of genes under stronger purifying selection and vice versa. Blue lines represent male stages and red lines represent female stages. The top and bottom limits of the boxes demarcate the s.d., based on 50,000 bootstraps. Life cycle stages (x axis) are denoted as follows: GA, gametophyte; PSP, parthenosporophyte; Early, early (2–5 cell) stages; Imm, sexually immature stages; Mat, sexually mature stages.

Full size image

In addition, we computed the average purifying selection acting on each stage through the transcriptome divergence index (TDI) metric, where a lower TDI indicates a transcriptome composed of genes under stronger purifying selection and vice versa (Methods, Supplementary Table 5). Compared with the TAI data, we observed that gametes, but not spores, consistently exhibited a high TDI (Fig. 4b). Our observations suggest that whereas evolutionarily young genes are more likely to be expressed in unicellular stages, genes under relaxed purifying selection are disproportionately found in gametes.

The TAI profile was markedly different between males and females during multicellular development (Fig. 4a and Extended Data Fig. 6). Ectocarpus males had higher TAI values than females during the early parthenosporophyte stage, whereas the reverse occurs during the immature gametophyte and parthenosporophyte stages. Of note, the sexual difference in TAI during the gametophyte development culminated at the immature stage, which is the stage when the gametophyte transcriptomes are most different between males and females57.

We further focused on the genes that underlie the high TAI values in unicellular stages. Based on the partial TAI value of each individual gene, \({{\rm{pTAI}}}_{i}={{\rm{ps}}}_{i}\times {e}_{is}/\,{\sum }_{i=1}^{n}{e}_{is}\) (where eis denotes the expression level of a given gene i at stage s, psi is its gene age assignment, and n is the total number of genes)41, we identified 115 genes in males and 98 in females that most strongly contributed to the TAI across all unicellular stages (Extended Data Fig. 9) (Methods). Gene Ontology (GO) term analysis did not retrieve any functional enrichment, especially since fewer than 10% of these genes have functional annotation. By contrast, the same analysis using genes with the strongest contribution to the TAI in the multicellular stages returned older genes with GO term enrichment linked to translational, organellar and transcriptional processes in both males and females (Extended Data Fig. 10 and Supplementary Table 6).

Finally, to broaden the species diversity and thus further test the generality of transcriptome conservation during multicellular development in brown algae, we inferred gene age (Extended Data Fig. 1) and computed TAI in the kelps L. digitata and S. polyschides (using 16,298 and 15,030 genes, respectively) (Supplementary Table 2 and Methods). Both species alternate between morphologically complex sporophyte and morphologically simple filamentous gametophyte generations (Extended Data Figs. 11 and 12). In the sporophyte development of L. digitata, we observed the lowest TAI during an embryo stage, whose transcriptome is closest to the most conserved stages in Fucus (Extended Data Fig. 12). Meanwhile, in the kelp gametophyte development, we observed a lower TAI during multicellular stages compared to unicellular samples (Extended Data Fig. 12), similar to Ectocarpus. In sum, whereas in the morphologically complex species or generations, the period of transcriptome conservation was coupled to the transition from cell differentiation to proliferation during embryogenesis, in the morphologically simple Ectocarpus as well as in the gametophytes of L. digitata and S. polyschides, the switch to multicellularity from unicellular stages constrains transcriptome evolution.

Discussion

The developmental hourglass model has been reported (and debated) at the molecular level in animals5,6,7,8,21,22,23,24,25,28, plants9,27,43,45,56,58, and fungi10,59,60. Here we examined the prevalence of the transcriptomic hourglass pattern in an independently evolved complex multicellular lineage, the brown algae.

A transcriptomic hourglass in Fucus

We show that embryogenesis of morphologically complex brown algae is underpinned by a transcriptome evolution pattern that is consistent with an hourglass model of embryonic evolution. Fucus embryos were more divergent at the earliest and latest stages of embryogenesis but presented a more conserved transcriptome during the mid-embryonic period, which serves as a body plan blueprint for the adult organism.

The hourglass pattern in Fucus was associated with reduced expression of evolutionarily young genes, rather than the upregulation of evolutionarily older genes, during the conserved mid-embryonic period, analogous to observations in animals and plants5,9,24,45. This pattern, together with data from kelp sporophytes, suggests that ancient genes (rather than genes specific to brown algae) form the Fucus and kelp body plans, pointing to a cis-regulatory hypothesis for the redeployment of pre-existing genes in the evolutionary innovations associated with brown algal embryogenesis61. Moreover, more broadly expressed, potentially pleiotropic, genes were associated with the evolutionarily conserved stages of development in Fucus, mirroring animal embryogenesis46,47. Notably, the waist of the hourglass corresponded to a major ontogenetic transition, from a cell-type differentiation stage where the body plan of the adult Fucus is established, to a proliferative stage, where development is largely characterized by growth of established cell types. This observation is reminiscent of the transition from primitive development to definitive development62,63 and is consistent with the ‘organizational checkpoint’ hypothesis, which postulates that a major transcriptome switch during transitions from cell fate acquisition into multicellular growth phases underlies the transcriptomic hourglass seen in animals, plants and fungi18.

The evidence that we present for the hourglass model, particularly the evolutionary novelty in the early stages, follows a lineage of studies on the establishment of polarity in brown algal zygotes64, which details variability in early embryogenesis. For example, the role of maternal cytoplasmic contribution and extrinsic cues differ markedly between Fucus65,66, Dictyota67 and Saccharina68. These observations are consistent with the variations that are also seen in early animal and plant development, which is followed by a more conserved period in mid-embryogenesis69. We suspect that the hourglass model also describes development in morphologically complex brown algae beyond the transcriptomic level.

Young genes in reproductive tissues

Transcriptome profiling of adult Fucus tissues revealed that male reproductive organs express a younger transcriptome, suggesting that evolutionarily young genes are more permissively expressed. This pattern is likely associated with the presence of male gametes (sperm) in the reproductive organs of male individuals. These observations complement independent findings in animals and plants where young genes are enriched in male reproductive tissue51,52,53,54,55,56. The younger transcriptome of Fucus sperm may be associated with sexual selection, which acts mainly through gamete-level interactions in sessile broadcast spawners such as Fucus70. Furthermore, the expression of younger genes in Fucus males is consistent with recent findings in the brown alga Macrocystis, implicating the female development programme as the morphogenetic ‘ground state’, superimposed upon in males71,72, though the mechanism behind this pattern in non-reproductive tissues is unclear.

By contrast, Ectocarpus did not exhibit differences in transcriptome age in male compared to female gametes. This result is consistent with the low level of sexual dimorphism typical of near-isogamous species57,73, limiting the extent of sexual selection in males compared to females, as well as the smaller set of sex-biased genes in Ectocarpus compared to Fucus and Macrocystis74.

Multicellularity constrains transcriptome evolution

Unlike Fucus, Ectocarpus development consists of two independent, multicellular, morphologically simple life stages connected by three types of unicellular stages (gametes, meiospores and mitospores). Crucially, the full life cycle can proceed without a canonical embryogenesis36. Although we did not find a classical ‘hourglass’ signature in Ectocarpus, we did find that multicellular developmental stages exhibit more conserved transcriptomes compared to unicellular stages. This finding was also supported in the gametophytes of two kelp species, L. digitata and S. polyschides. Furthermore, the low TDI associated with multicellular stages of development likely reflects ongoing purifying selection, implicating multicellularity per se as under evolutionary constraint. This evolutionary pattern in filamentous life stages may be due to the lack of a singular ‘mid-embryonic period’, where the body plan is established. Instead, Ectocarpus (as well as L. digitata and S. polyschides gametophytes) exhibits a ‘modular’ development, in which cell types are differentiated continuously over time, reiterating a filamentous body plan. We reason that the regulatory complexity from cell-type differentiation programmes activated across filamentous multicellular stages results in an overall conserved transcriptome. We posit that these constraints may have become concentrated towards a more singular ‘mid-embryonic period’ during the evolution of lineages with increased morphological complexity, such as Fucus.

In conjunction with the pattern of ‘multicellularity constraint’, unicellular dispersal stages may be more permissive to evolutionary change75. Differences in the cell structure of unicells, such as the lack of a cell wall, can result in (a)biotic exposure which opens new selection opportunities76 compared with multicellular stages. For example, it is well known that virus infection occurs in the unicellular stages (gametes and spores) in Ectocarpus77. Of note, whereas TAI was similarly high for all unicellular stages, only gametes show decreased purifying selection, suggesting a signal for sexual evolution at the sequence substitution level that is specific to gametes. Our results demonstrate that species that lack multicellular organization via embryogenesis may still exhibit a developmental window with higher transcriptome conservation compared to unicellular stages.

Together, we present evidence for the existence of a developmental hourglass pattern during embryogenesis in morphologically complex brown algae, analogous to hourglass patterns previously reported in animals, plants and fungi. Our distinction between complex multicellular development and embryogenesis suggests that transcriptome conservation patterns are a fundamental characteristic of complex multicellularity itself, with possible downstream effects on embryogenesis.

Methods

Sample preparation

Details of algal strains used are described in Supplementary Table 1. Fucus embryos were prepared as described previously78. In brief, gametes were allowed to release in 2 l natural seawater (NSW), and cleaned using several NSW baths for female gametes and phototaxy for male gametes. Then, gametes were mixed together for 1 h and fresh zygotes were cleaned as for female gametes. F. serratus and F. distichus embryos grow highly synchronously and at least 10,000 developing embryos were flash frozen at specific developmental stages (see Extended Data Fig. 5). Embryos were grown at 14 °C in NSW (F. serratus) or at 10 °C in diluted NSW (F. distichus) supplemented, for one week, with GeO2 (0.4 mg l−1) to avoid growth of diatoms. Media was changed weekly. For both species of Fucus, embryos were grown under neutral day conditions (12 h/12 h day/night cycle). Ectocarpus life cycle stages were grown at 14 °C in Provasoli-enriched NSW (PES)79 a 12 h/12 h day/night cycle and 20 µmol photons m−2 s−1 irradiance, as described previously80.

Kelp gametophytes were grown vegetatively in 50% PES with iodine enrichment79,81 at 14 °C under red light in a 14 h/10 h day/night cycle. To induce fertility, gametophyte tufts were carefully ground using a mortar and pestle and the gametophyte fragments were sown at low density (approximately 500 gametophytes per cm2). Gametophyte maturity was induced at 12 °C and 20–30 µmol photons m−2 s−1 white light under a 16 h/8 h day/night cycle in full-strength PES. Fertile gametophytes with visible gametangia were collected after seven days. Both sexes were co-cultivated to induce maturity and sporophyte production. Field samples were collected from Santec (France), Perharidy (France) and Kiel (Germany) as described in Supplementary Table 1.

Note that Fucus and Ectocarpus are broadcast spawners, releasing their gametes in the surrounding seawater, where subsequent development takes place. Early development can therefore be followed in a large number of replicate individual clones that develop highly synchronously, without potentially contaminating parental tissue, greatly facilitating experimental approaches. Meanwhile, in L. digitata and S. polyschides, access to biological material is challenging, resulting in a less complete dataset for these species. For example, obtaining sufficient quantities of very early sporophyte stages of L. digitata was impeded by their attachment to the maternal gametophyte.

RNA extraction from brown algae

Adult tissues of Fucus and kelps were quickly brushed and then rinsed with filtered and autoclaved NSW. Different parts such as holdfast, stipe, meristem, vegetative tissue and reproductive tip were sliced into 0.5–1 cm pieces and transferred into 1.5 ml low-bind Eppendorf tubes. The tubes were snap frozen in liquid N2 and the stored at −80 °C until further processing, as done for other stages of Fucus and kelp development as well as Ectocarpus.

The RNA extraction protocol followed previous publications71,82 and is described in Supplementary Table 1. Snap-frozen algae were dry-ground with a pestle in liquid N2/dry ice and mixed with 750 μl of freshly prepared RNA-extraction buffer (100 mM Tris-HCl pH8.0 (Thermo Fisher AM9856); 1.4 M NaCl; 2% CTAB (Sigma Aldrich, 52365-50G); 20 mM EDTA pH8.0; 1% β-mercaptoethanol; 2% polyvinylpyrrolidon (Thermo Fisher AM9690)) preheated to 65 °C. Then 250 μl 5 M NaCl were added into the tubes. An equal volume of chloroform: isoamylalcohol (24:1) was added and mixed well followed by centrifugation at 10,000g for 15 min at 4 °C. The aqueous phase was removed into RNAse-free tubes and extracted again with 250 μl pure ethanol and an equal volume of chloroform: isoamylalcohol (24:1) as before. RNA was precipitated by adding LiCl (Thermo Fisher, AM9480) to a final concentration of 4 M, together with 1% volume of β-mercaptoethanol, mixing and incubating at −20 °C overnight.

The RNA was pelleted by centrifugation at full speed (>18,000g) for 45 min to 1 h at 4 °C. RNA was washed with 70% cold ethanol and the pellet was air dried for 3–5 min and then the RNA was dissolved in 30 μl RNAse-free H2O. Residual DNA was eliminated using the TURBO DNase Kit (Thermo Fisher, AM1907) according to the manufacturer’s instructions. The final RNA concentration and size distribution were determined using a Qubit (RNA BR Assay Kit, Invitrogen, Q10210) and an RNA Nano bioanalyzer (Agilent, 5067-1511).

RNA-seq

The RNA-seq libraries were prepared using commercially available kits according to the manufacturer’s instructions. Poly-A selection for mRNA enrichment was performed using the corresponding NEB kit (E7490) followed by library preparation using the directional RNA library prep kit from NEB (E7765). A Single Cell/Low Input RNA library prep kit (NEB, E6420) was used to synthesize additional cDNA and prepare sequencing libraries for samples where it was not possible to obtain large amounts of material (meiospores and mitospores in Ectocarpus; earlySP in L. digitata; matGA in S. polyschides) (Supplementary Table 1). The Qubit 1× dsDNA HS assay kit (Invitrogen, Q33230) was used to determine the final DNA concentration of the libraries and the DNA high-sensitive Kit (Agilent, 5067−4626) was employed for bioanalyzer analysis to evaluate the distribution of insert sizes.

Sequencing was performed on a NextSeq2000 instrument with sequencing kit P3-300 (Illumina). The libraries were pooled for sequencing such that for each library we obtained about 30,000,000 reads, corresponding to 9 Gb of data (Supplementary Table 1).

RNA-seq datasets were processed using the nf-core/rnaseq pipeline v3.583,84. For all three species, expression quantification was performed using salmon v1.5.285, to ensure consistency, and imported to R using tximport v1.26.186. For Ectocarpus, RNA-seq reads were pseudo-mapped to transcripts inferred for each gene from version 2 of the Ectocarpus species 7 genome11. For L. digitata and S. polyschides, the same method was applied to recently published genomes12. Previously published S. polyschides immGA data71 was remapped to maintain consistency. Since high quality public genomes were not available for F. serratus and F. distichus, the quantification was carried out on recently published de novo transcriptome assemblies74.

We precluded genes with mean length-scaled TPM (transcripts per million) across samples below 2 from subsequent analyses. In analyses indicated as ‘denoised’, we further removed genes with noise-like behaviour using the ‘counts’ mode of noisyR v1.0.044.

Transcriptome age index

The TAI captures the average gene age of a given transcriptome, weighted by the expression level of each gene5,41. The relative age of each gene in Ectocarpus, F. serratus, F. distichus, L. digitata and S. polyschides was inferred using GenEra v1.042, based on genomic phylostratigraphy87. In brief, GenEra takes all protein-coding genes and pairwise aligns these sequences against the taxonomy-resolved NCBI non-redundant database88,89, using DIAMOND v2.0.14 (‘sensitive’ mode; e-value < 10−5)90. Next, search hits are filtered by their distribution across taxonomic nodes until the most distant taxonomic node is determined as the ‘gene age’ (or removed as potential contamination), with the evolutionarily oldest genes assigned as phylostratum (PS) 1 and the youngest assigned as PS 8 in F. serratus, F. distichus and S. polyschides, PS 10 in L. digitata and PS 11 in Ectocarpus. PS 7 corresponds to the origin of brown algae (complex multicellularity). For genes with more than one isoform, the age of the oldest isoform was used. Thus, after filtering lowly expressed genes across all samples (TPM < 2) and potential contaminations, we obtained expression and evolutionary data for 8,291 genes in F. serratus, 7,907 genes in F. distichus and 11,571 genes in Ectocarpus. Using myTAI v1.0.1.900041, TAI was calculated for each stage (TAIs) as follows,

$${{\rm{TAI}}}_{s}=\mathop{\sum }\limits_{i=1}^{n}\left(\frac{{{\rm{ps}}}_{i}\cdot {e}_{is}}{\mathop{\sum }\limits_{i=1}^{n}{e}_{is}}\right)$$

where psi denotes the relative gene age (phylostratum) for a given gene i. The term eis denotes the expression level of a given gene \(i\) at developmental stage s and n denotes the total number of genes.

The expression level was captured using TPM values, since we are quantifying the relative abundance of mRNA molecules per gene rather than the count of sequencing fragments. To test the stability of the TAI profiles and reduce the variance in the highly expressed genes23,43, we performed several RNA-seq data transformations on the expression matrices: square-root transformation (used for the main figures), log transformation with a pseudo-count of 1 (log2(TPM + 1)), ‘regularized log’ transformation91 (rlog), and rank transformation (that is, genes were ranked by level of expression at each stage). To reduce potential outliers, the median abundance value of replicates was chosen to represent the expression level (eis).

The statistical significance of the resulting profiles was assessed using non-parametric permutation tests (flat line test, reductive hourglass test and one-sided pairwise TAI test), using the FlatLineTest(), ReductiveHourglassTest() and PairwiseTest() functions implemented in myTAI41. The P value defines (for each tested shape) the probability that the observed TAI pattern is drawn from a random set of TAI profiles with permuted gene ages. We defined ‘early’ stages as S1–2, ‘mid’ as S3–4, and ‘late’ as S5 in F. serratus, and ‘early’ stages as S1–4, ‘mid’ as S4.5, and ‘late’ as S5 in F. distichus, due to differences in developmental stage correspondence. These tests, including those for sex differences, were performed with 50,000 permutations.

For the pTAI analysis, we used the function pMatrix() implemented in myTAI41, which calculates the contribution of each gene to the TAI at each stage by multiplying the phylostratum of each gene by its expression level divided by the sum of expression of all genes, that is,

$${{\rm{pTAI}}}_{is}=\frac{{ps}_{i}\cdot {e}_{is}}{\mathop{\sum }\limits_{i=1}^{n}{e}_{is}}$$

where, like TAI, psi denotes the relative gene age (phylostratum) for a given gene i and eis denotes the expression level of a given gene i at developmental stage s and n denotes the total number of genes. The elbow method was used to identify 500 genes with the highest TAI contribution in each developmental stage; genes driving the TAI value across all unicellular or multicellular stages were inferred via intersection. For consistency with the main TAI analyses, square-root transformation was applied before the pTAI analysis.

Transcriptome specificity index

To investigate whether the transcriptome at the waist of the hourglass is composed of broadly expressed genes compared to other stages, we first indexed each gene by its relative expression specificity/breadth across development using tau48,49, that is,

$${{\rm{tau}}}_{i}=\frac{\mathop{\sum }\limits_{i=1}^{N}\left(1-{\widehat{e}}_{i}\right)}{N-1};{\widehat{e}}_{i}=\frac{{e}_{i}}{\mathop{\max }\limits_{1 < i < n}\left({e}_{i}\right)}$$

where N is the number of stages, \({\widehat{e}}_{i}\) is the expression level of a given gene i normalized by the maximal expression value. A lower tau indicates low stage-specificity (in other words, broad expression), and vice versa. The resulting tau values across all genes are stratified into deciles (tau-stratum), which enables analogous comparisons to TAI. It should be noted that PS and tau are not correlated (Kendall’s τ ≈ 0.05 in both Fucus species), indicating that these metrics capture independent signals. In contrast to the TAI, the TSI captures the average expression specificity/breadth of a given transcriptome, weighted by the expression level of each gene, that is,

$${{\rm{TSI}}}_{s}=\mathop{\sum }\limits_{i=1}^{n}\left(\frac{{{\rm{ts}}}_{i}\cdot {e}_{is}}{\mathop{\sum }\limits_{i=1}^{n}{e}_{is}}\right)$$

where tsi denotes the relative expression specificity/breadth (tau-stratum) for a given gene i, eis denotes the expression level of a given gene i at developmental stage s and n denotes the number of genes. The median abundance of replicates was chosen to represent the expression level (eis). Existing functions in myTAI were repurposed for this analysis.

Transcriptome divergence index

To explore whether unicellular stages in Ectocarpus not only exhibited a young transcriptome, but also genes under relaxed purifying selection, we computed the TDI. In contrast to the TAI, the TDI captures the average gene selective pressure (divergence-stratum; based on deciled dN/dS ratios) of a given transcriptome, weighted by the expression level of each gene. The divergence-stratum of each gene in Ectocarpus was inferred from dN/dS ratios using orthologr43. In brief, one-to-one orthologues were inferred between Ectocarpus sp. 7 and Ectocarpus subulatus92, using best reciprocal hits, and the dN/dS ratio was computed using the default “Comeron” estimation method. Importantly, >99% of one-to-one orthologue comparisons fell below the dN/dS ratio of 1, indicating that we are quantifying the degree of purifying selection. Next, the resulting dN/dS ratios across all genes are stratified into deciles, with the scale ranging from 1 (strong purifying selection) to 10 (weakest purifying selection). Fucus species were precluded from this analysis due to the short divergence time between F. serratus and F. distichus, approx. 4 million years ago93, resulting in more than 10% of genes having dN/dS of 0. For genes with more than one isoform, the divergence-stratum of the oldest isoform was used.

Using myTAI41, we calculated the TDI for each stage as follows,

$${{\rm{TDI}}}_{s}=\mathop{\sum }\limits_{i=1}^{n}\left(\frac{{{\rm{ds}}}_{i}\cdot {e}_{is}}{\mathop{\sum }\limits_{i=1}^{n}{e}_{is}}\right)$$

where dsi denotes the relative divergence level (divergence-stratum) for a given gene i. The term eis denotes the expression level of a given gene i at developmental stage s. The median abundance of replicates was chosen to represent the expression level (eis).

Distance/similarity-based transcriptome comparison

To quantify the overall distance/similarity between the transcriptomes of embryo stages in the two Fucus species, we computed the Pearson correlation, Spearman correlation, Manhattan distance and Jensen–Shannon distance (JSD) metric. Several metrics were employed owing to issues of calculating distance in high-dimensional data94. To compare expression levels between species, we compared the expression levels (abundance) of orthogroups (sets of orthologues and paralogues) using OrthoFinder v2.5.495, treating genes as isoforms and orthogroups as genes when importing the RNA-seq data using tximport v1.26.186. Note, expression was quantified using length-normalized TPM to avoid bias from different gene lengths between species. Orthogroups were used rather than one-to-one orthologues (inferred through procedures such as best reciprocal hit), since orthogroups also capture the expression profile of in-paralogues, thereby covering more genes in the genome. The abundance data was transformed using rlog91. Correlation matrices were computed using cor() from the stats package in R96, while Manhattan and JSD metrics were computed using the R package philentropy v0.7.097. We employed the same approach (using log-transformed orthogroup abundance) to compare the overall transcriptome distance/similarity between life cycle stages in Ectocarpus and embryo stages in the two Fucus species, as well as to determine the corresponding stages between an early embryo stage of L. digitata sporophyte and each embryo stage of Fucus.

Enrichment analyses

To explore gene function, GO terms were obtained using InterProScan v5.61-93.098. GO enrichment analysis was then performed on genes contributing most to Ectocarpus TAI (inferred from the partial TAI value of each individual gene, pTAIi) using Fisher’s exact test statistics with the parent–child algorithm as implemented in TopGO v2.48.099,100. Statistical tests and significance levels are indicated in the text and figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.