Animal multicellularity and polarity without Wnt signaling

Acquisition of multicellularity is a central event in the evolution of Eukaryota. Strikingly, animal multicellularity coincides with the emergence of three intercellular communication pathways – Notch, TGF-β and Wnt – all considered as hallmarks of metazoan development. By investigating Oopsacas minuta and Aphrocallistes vastus, we show here that the emergence of a syncytium and plugged junctions in glass sponges coincides with the loss of essential components of the Wnt signaling (i.e. Wntless, Wnt ligands and Disheveled), whereas core components of the TGF-β and Notch modules appear unaffected. This suggests that Wnt signaling is not essential for cell differentiation, polarity and morphogenesis in glass sponges. Beyond providing a comparative study of key developmental toolkits, we define here the first case of a metazoan phylum that maintained a level of complexity similar to its relatives despite molecular degeneration of Wnt pathways.


Results
Conserved Notch pathway. In contrast to the previous analysis of A. vastus transcriptome 22 , we identified the Notch ligand Delta in glass sponges. Indeed, consistent with the analysis of Oscarella carmela, a unique Delta protein was retrieved in glass sponge, whereas five were previously described in Amphimedon genome 23 . This suggests that a single ligand was certainly present in the last common ancestor of sponges and that other paralogs arose later by lineage specific duplications. Computational domain analyses allowed us to confirm the conservation of the overall structure of glass sponge Delta proteins (Fig. 2b). Indeed, the DSL domain, several EGF domains and the transmembrane region were predicted by Interproscan. In blue the components retrieved in glass sponges, in black the element absent in glass sponges and other poriferan and in red the component retrieved in other sponges but not in glass sponges. (b.) Phylogenetic analysis revealing three BMP-like, one TGFβ-like and one unclassified TGF ligand in glass sponges. (c.) Alignment of TGFβ-like motif. In blue the RXXR cleavage site. (d.) Phylogenetic analysis revealing three type I and two type II TGFβ receptors in glass sponges. (e.) Phylogenetic analysis revealing Smad1/5, Smad2/3 and Smad4 in glass sponges. (b.,d. and e.) Black, grey and white triangles represent clades of proteins with a node support higher than 95%, comprise between 70% and 95% or lower than 70%, respectively. (f.) Domain composition of TGFβ ligands and receptors. Intriguingly, the Notch-binding domain of Delta/Jagged proteins (i.e. MNLL) is not predicted in sponges except in one of the five proteins retrieved in Amphimedon 23 . However, this protein does not show a typical composition, as it lacks EGF motifs. In this context, we draw attention on a pattern of cysteines conserved between MNLL and the N-terminus of all poriferan sequences (Fig. 2c). Since cysteines are extremely conserved only when involved in functional interactions 24 , it likely reflects the conservation of the MNLL. As we found bona fide Notch proteins as well (i.e. several EGF, Notch domains, one transmembrane domain, five ANKR and at least one PEST domain) (Fig. 2b), it appears that Delta/Jagged activation and signal transduction through the membrane may occur in glass sponges. Functional studies are however required to prove ligand/ receptor interactions in poriferans and to identify key residues.
We also identified in glass sponges all the proteins involved in the multiple cleavages of the Notch receptor. We retrieved Furin (Fig. 2d), two ADAM10/17 proteins (Fig. 2d and e) and the four ɣ-secretase subunits (i.e. Presenilin, Presenilin Enhancer 2, Anterior Pharynx defective I and Nicastrin) (Supplementary Table 1). Interestingly, these proteins, required for S1, S2 and S3 cleavages respectively, show a similar domain composition to their human orthologs ( Fig. 2d and f). Thus, we assume that Notch processing and the release of the NICD signaling fragment is possible in glass sponges.
Finally, glass sponges also possess the CSL/SuH transcription factor and a protein sharing similarities with the NCoR/SMRT co-repressor (Fig. 2g), whereas in agreement with previous analyses of the pathway in Demospongiae and Homoscleromorpha 23 , no MAML protein was retrieved. Thus, all the core proteins required for Notch pathway activity are present in glass sponges (Fig. 2a).
Conserved TGFβ pathway. By dissecting the diversity of the TGFβ core components in early branching metazoans (i.e. Nematostella, Mnemiopsis, Trichoplax and Amphimedon) ( Table 1), previous studies revealed that the number of TGFβ receptors and Smad proteins is relatively stable in these lineages, whereas the number of ligands is highly variable 25 . The genome of Amphimedon however challenges this statement by revealing many Smad paralogs. In addition, many TGFβ ligands retrieved in Amphimedon and Mnemiopsis are still unclassified. Thus, the number of ligands and Smad proteins in the last common ancestor of Metazoa and their evolution throughout animal kingdom remains difficult to assess as poriferans and ctenophores represent the earliest branching metazoan phyla 15,[26][27][28] .
We found five TGFβ ligands in glass sponges ( Fig. 3b and Supplementary Table 2). One appears to belong to the TGFβ-like sub-family (i.e. TGFβLa) while three others are clustered within BMP-like proteins (i.e. BMPLa, BMPLb and BMPLc). The last one branches basally, making it difficult to assign it to any TGFβ subgroup. Interestingly, the close examination of each sub-family revealed that TGFβ-like proteins show a C-Xn-CC motif (where n is comprised between 6 and 8) between the cleavage site and the mature peptide, whereas BMP-like proteins show a unique cysteine confirming the assignment of glass sponge BMPs to this sub-family (Fig. 3c). In contrast, the TGFβ-like motif was found in TGFβLa and in the fifth member, which we propose to name TGFβLb.
Since the TGFβ-like cysteine motif described here appears as an efficient tool to shed light on the evolution of TGFβ ligands when phylogenetic analyses are not informative, we tried to confirm the assignment of Amphimedon and Mnemiopsis proteins and to decipher whether the unclassified proteins 25 belong to the BMP-like or the TGFβ-like subfamilies. Through this approach, we suggest that all previously unclassified proteins are BMP-like proteins ( Table 1 and Supplementary Figure 1). Our analysis reveals that sponges and ctenophores possess several BMP-like proteins and only two TGFβ-like proteins, likely reflecting the situation in the last common ancestor of animals. This finding calls for functional studies to examine the role of these cysteines, which were subjected to a very high pressure of selection across animal evolution.
In addition to TGFβ ligands, three type I receptors (TgfR Ia, TgfR Ib, ActR I), two type II receptors (TgfR II and ActR II) and three Smad proteins (Smad2/3, Smad4 and Smad1/5) were identified (Fig. 3d and e), whereas like in the Amphimedon genome 25 we failed to find Smad6/7, SARA, the co-repressor Ski/Sno, but also SMURF in either transcriptomes. Thus, although the number of Smad1/5 paralogs and the presence of Smad6/7 in the last common ancestor of animals remain difficult to assess, the analysis of glass sponge transcriptomes allows us to propose a unique copy of Smad4 and Smad2/3 at the dawn of animals, while several duplications may have occurred in Demospongiae. At the protein level, domain analyses revealed that despite the absence of the propeptide in most glass sponge TGFβ ligands (Fig. 3f), they all possess the characteristic RXXR cleavage site (except OmBMPLb) (Fig. 3c) that would allow the release of the predicted mature peptide and thus would activate TGFβ receptors. Most receptors display a typical domain composition, consisting in a signal peptide, a transmembrane domain (except OmActRII), a GS domain (in type I receptors only) and a kinase domain (Fig. 3f), while the TGFβ-binding domain of the receptors is highlighted by a conserved cysteine pattern (Fig. 3f and Supplementary Figure 2). In addition, consistent with Smad proteins in other animals, the MAD homology 1 and the SMAD domain-like were found in all Smad proteins retrieved in glass sponges (Data not shown). Likewise, the DNA binding residues were predicted in Smad1/5 and Smad4 (but not in Smad2/3). In conclusion, all the elements (i.e. proteins and functional domains) necessary to transduce TGFβ signals into the nucleus are present in glass sponges, although some accessory proteins are missing (Fig. 3a).
Absence of key components of the Wnt pathway. Although conserved Notch and TGF-β pathways were retrieved in glass sponges, the situation was markedly different when considering the Wnt pathway since several key genes were entirely absent in both A. vastus and O. minuta. More specifically, we retrieved apc, axin, β-catenin, frizzled B (fzdB), groucho, gsk3, LRP, porcupine (porc; only in O. minuta) and tcf/lef but failed to identify frizzled A (fzdA), dishevelled (dvl), wnt and wntless (wls) (Supplementary Table 3).
In order to ensure that our observations faithfully reflect the glass sponge genome and not some technical issues, we searched for these 13 genes in various transcriptomes available for other sponge lineages: Homoscleromorpha (i.e O. carmela and O. sp), Calcarea (i.e. L. complicata and S. ciliatum) and Demospongiae (i.e. E. muelleri). We unambiguously identified these proteins in all transcriptomes ( Table 2 and Supplementary  Table 4-6) and confirmed our findings by a KAAS analysis (Supplementary Figure 3) [29][30][31] . Consequently, we assume that the absence of Dvl, FzdA, Wls and Wnt in both glass sponge transcriptomes likely reflects their absence in their genomes.
Beyond these four missing genes, five others encode proteins that lack key functional domains. Groucho, GSK3, LRP and Porc show a regular domain composition compared to their human orthologs (Fig. 4b,c), whereas the others (APC, Axin, β-catenin, FzdB and TCF) do not. The region known to interact with β-catenin is not predicted in Axin proteins found in glass sponges (Fig. 4d), and the APC protein is truncated, lacking the region allowing β-catenin, GSK3 and Axin binding (Fig. 4e). This is consistent with previous analyses of Axin in demosponges (i.e. Amphimedon queenslandica), and APC proteins from demosponges and cnidarians (e.i. Nematostella vectensis) 17 . Thus, the formation of the β-catenin destruction complex (i.e. Axin, APC and GSK3) is questionable in non-bilaterian animals, especially when considering the absence of Axin protein in the genome of the calcarean Sycon ciliatum 32 . Supporting this idea, Schippers and Nichols succeeded to identify adherent junction components (i.e. α-catenin and classical cadherin) but not Axin, APC or GSK3 as Emβcatenin binding partners by immunoprecipitation 33 . Nevertheless, chemical inhibitions suggest that Emβcatenin can be considered as a phosphorylation substrate of GSK3. Thus whether this phosphorylation involves the two other components and drives β-catenin degradation still requires further investigation in basal metazoans. The domain composition of FzdB, β-catenin and TCF is also unusual in glass sponges, even when compared to their demosponge orthologs: i) the Dvl-binding motif (KTXXXW) is absent/divergent in FzdB (Fig. 4f and g), ii) the β-catenin shows a divergent GSK3-binding motif (Supplementary Figure 4), and lacks the Dvl-binding domains (i.e. PDZ-binding motif: S/ TxV/L) (Fig. 4h), iii) the β-catenin-binding domain of TCF is not predicted (Fig. 4h). The absence of three core proteins and the loss of some key functional domains or motifs in the remaining components support the idea that the Wnt pathway cascade cannot be considered as functional in syncytial glass sponges. We nevertheless envisage that the activation of Wnt pathway responsive genes may occur in glass sponges, but in a Wnt-independent way since the transcription factor Tcf/Lef and its co-repressor Groucho were characterized here.

APC Axin B-cat
Analyses of the CEGs. We previously described the first transcriptome of Oopsacas minuta 34 . Here we evaluated the completeness of several transcriptomic databases available not only for glass sponges (i.e. Aphrocallistes vastus) but also for other poriferan lineages (i.e. Ephydatia muelleri, Leucosolenia complicata, Oscarella carmela, Oscarella sp. and Sycon ciliatum) by searching the previously defined 248 core eukaryotic genes (CEGs) [35][36][37][38] . Despite high variability in the total number of base pairs among sponges (from 28.1 Mb to 68.6 Mb) ( Table 3), a remarkably high number of CEGs (from 87.5% to 96.8%) was nonetheless recovered for most studied species (Supplementary Figure 5). Given the high percentage of CEGs present in each poriferan database, the collected datasets can be considered of high quality and that comparison between them is meaningful. Finally, only five CEGs absent in O. minuta (out of 14) are also missing in A. vastus, suggesting that taken together, these two transcriptomes likely reflect the overall genome of glass sponges.

Discussion
Almost ten years ago, the acquisition of the first genome of sponges shed light on the conserved molecular toolkit of animal development, revealing that despite their apparent anatomical simplicity (i.e. few cell types and no organs), sponges share key developmental signaling pathways (i.e. Notch, TGFβ and Wnt modules) with more complex animals 6 . While the molecular characterization of these three modules has been investigated in Amphimedon queenslandica 17,23,25 , comparative approaches in the other three poriferan lineages are still sparse 22,23,25,32 , especially for the deep sea phylum of glass sponges 22 . Briefly, our in depth analysis of two glass sponge transcriptomes supports that TGFβ and Notch signaling are likely functional in glass sponges. Indeed, we retrieved at least the minimal set of components required to transduce Notch and TGFβ signals from the membrane to the nucleus 39 , confirming the existence of both signaling in the last common ancestor of animals 6,23 . The present study also uses various transcriptomic databases to reveal the lack of the upper part of the Wnt signaling cascade in glass sponges.
In theory, it remains possible that our data reflect the absence of several Wnt pathway transcripts at the moment of sampling. This is, however, unlikely for multiple reasons: i) the analysis of CEGs allowed us to be highly confident regarding the coverage of the databases used in this study; ii) the exact same set of genes appeared to be absent in two independently acquired transcriptomes (i.e. Aix Marseille University, France and University of Alberta, Canada), while complete Notch and TGF-β modules were retrieved; iii) several paralogs of wnt and dvl genes (i.e. 3 to 21 wnt and 1 to 2 dvl) were previously described in other sponge lineages. Among them, some are dedicated to embryogenesis while some others are highly expressed in adult under homeostatic conditions 17,32,40 . Thus, by collecting adult specimens that contained numerous embryos at all developmental stages 41-43 , we should have easily retrieved at least some of these paralogous genes like it was the case for all other transcriptomes; iv) only a fraction of Wnt pathway gene transcripts were missing, while others were retrieved, suggesting that sampling did not skew our analysis.
Although the loss of a bona fide Wnt pathway in glass sponges clearly coincides with their atypical syncytial organization, it remains unclear whether both features are linked. Leys has argued that cytoplasmic streaming within the trabecular syncytium and through the pores of plugged junctions (Fig. 1e) may cause efficient but localized trafficking, allowing signals to be transmitted 19 . Thus, the syncytial organization of glass sponges could support signal transmission, making paracrine Wnt signaling through cell membranes dispensable. It is also interesting to note that the syncytial organization of glass sponges occurs following the fifth division (32-cell stage), when cells massively reorganize in a coordinated way 20 . In other animals, such as cnidarians and mammalians, the 32-cell stage marks the onset of Wnt pathway activation 44,45 . Therefore, the two outstanding features of glass sponges -the lack of a proper Wnt pathway and of a fully cellular organization -are most likely deeply interwined. We hypothesize that in this phylum, the acquisition of a syncytial organization with plugged junctions  may have allowed an alternative mode of cell-to-cell communication during morphogenesis, thus bypassing the paracrine Wnt pathway and enabling its partial loss in the last common ancestor of glass sponges. Our results, together with recent genomic and transcriptomic data on myxozoans 46 -highly derived endoparasitic cnidarians devoid of Wnt pathway -, challenge the proposed universal importance of Wnt pathways in early development of animals 47 . Indeed, these unexpected findings suggest that the Wnt core (i.e. Wnt ligand secretion and Dsh translocation to the cell membrane) is needed neither to reach a multicellular stage, nor for cell differentiation and cell polarity (e.g. collar body of glass sponge; capsulogenic cells of myxozoans), nor for the acquisition of body axis polarity (Fig. 1h and i). In contrast with the drastically simplified morphology of myxozoans with respect to the typical body plan of cnidarians, glass sponges have maintained a body plan comparable to other sponge lineages, making comparative approaches more relevant and easier to carry out. It is, however, striking to note that in both lineages the absence of the Wnt core coincides with the acquisition of a syncytial organization 19,48 , suggesting that the Wnt pathway is required only in truly multicellular organisms. These peculiar exceptions across the metazoan tree strengthen the idea that the Wnt pathway has played a key role in cell-to-cell communication during the acquisition of animal multicellularity.
In conclusion, despite the fact that animal multicellularity is related to the acquisition of three main developmental signaling pathways, only the TGFβ and Notch modules were conserved in glass sponges, whereas the transition to a syncytial organization may have caused the erosion of the Wnt pathway. In the future, it will be necessary to study the molecular basis of plugged junction establishment and whether these junctions allow transportation and exchange of signaling molecules.

Methods
Samples for microscopy. Specimens of O. minuta were fixed in a mixture of 2% osmium tetroxide, saturated with mercury chloride (6/1), desilicated in 5% hydrofluoric acid 43 . Thin sections were contrasted with 0.5% uranyl acetate and lead citrate and viewed in a Hitachi Hu 600 transmission electron microscopes. For scanning electron microscopy, specimens were fractured in liquid nitrogen after desilification, sputter-coated with gold palladium and observed under a Hitachi S-570 scanning electron microscope.
Identification of orthologous genes. The Notch, TGF-β and Wnt pathway components were searched using a TBLASTN approach. To do so, sequences of Mus musculus were used as initial query and a 1.0 threshold e-value was applied in order to retrieve even highly divergent sequences. When no orthologous sequence was obtained, proteins already described in cnidarians and sponges were used as query. Orthologous assignations of candidate genes were done i) by reverse TBLASTN approach against NCBI database, ii) phylogenetic analyses and iii) domain analyses using Interproscan 5, epestfind as well as domain described in the literature 49,50 . In order to confirm our data concerning the Wnt pathway, we proceeded to an additional KEGG Automatic Annotation (http://www.genome.jp/tools/kaas/) [29][30][31] .
The 248 Core Eukaryotic Genes. A reciprocal best hits approach has been used to identify the 248 CEGs (Core Eukaryotic Genes) 35 in sponge databases (Supplementary Table 7). To do so, the 248 cds of human CEGs 35,51 were identified by TBLASTN (10E-100 threshold e-value) in the following database: Homo_sapiens.GRCh38.cds. all.fasta (ftp://ftp.ensembl.org/pub/release-80/fasta/homo_sapiens/cds/). The 248 human cds were used to identify candidates in sponges (TBLASTN, 10E-5 threshold e-value). Best hit were validated by a reciprocal approach against human (TBLASTN, 10E-5 threshold e-value). To do so, the 6 ORFs were generated for each poriferan candidate gene and the longest one was retained and used as query. Poriferan candidate genes were considered as orthologs to CEGs if the reverse best hit corresponded to the initial human cds or to a similar human sequence (coverage higher than 99% and e-value of 0.0). This allowed us to retrieve a large set of CEGs. Finally, phylogenetic analyses (alignment using ClustalW; ML analysis using PhyML 3.0, LG model and 1,000 bootstraps) and sequence identity matrixes (Bioedit software) allowed us to detect human in-paralogs (Supplementary Table 8) and thus to slightly increase the number of CEGs retrieved in each poriferan transcriptome. Data Availability Statement. All sequences described in the present study were deposited on NCBI website and the corresponding accession numbers are provided in the Supplementary Information file while full alignments and phylogenetic analyses are available upon request.