Evolutionary origin of gastrulation: insights from sponge development

Background The evolutionary origin of gastrulation—defined as a morphogenetic event that leads to the establishment of germ layers—remains a vexing question. Central to this debate is the evolutionary relationship between the cell layers of sponges (poriferans) and eumetazoan germ layers. Despite considerable attention, it remains unclear whether sponge cell layers undergo progressive fate determination akin to eumetazoan primary germ layer formation during gastrulation. Results Here we show by cell-labelling experiments in the demosponge Amphimedon queenslandica that the cell layers established during embryogenesis have no relationship to the cell layers of the juvenile. In addition, juvenile epithelial cells can transdifferentiate into a range of cell types and move between cell layers. Despite the apparent lack of cell layer and fate determination and stability in this sponge, the transcription factor GATA, a highly conserved eumetazoan endomesodermal marker, is expressed consistently in the inner layer of A. queenslandica larvae and juveniles. Conclusions Our results are compatible with sponge cell layers not undergoing progressive fate determination and thus not being homologous to eumetazoan germ layers. Nonetheless, the expression of GATA in the sponge inner cell layer suggests a shared ancestry with the eumetazoan endomesoderm, and that the ancestral role of GATA in specifying internalised cells may antedate the origin of germ layers. Together, these results support germ layers and gastrulation evolving early in eumetazoan evolution from pre-existing developmental programs used for the simple patterning of cells in the first multicellular animals.


Background
It has been long argued and widely accepted that early metazoan evolution included the progressive addition and elaboration of cell layers, with the first animals being little more than a small sphere of epithelial cells. The addition of a second inner layer is thought to have followed, resulting in the so-called 'gastraea' that possessed an ectoderm and endoderm. A middle mesodermal layer is proposed to have evolved last [1,2]. It has been noted that this progression is reflected in contemporary early branching animals, with sponges and placozoans being interpreted as having one, two or no germ layers, cnidarians and ctenophores possessing two or possibly three germ layers and bilaterians having three or possibly four [2,3]. The generation of two or more germ layers by gastrulation and the consistency of the fate of these layers unites the Eumetazoa (though exceptions exist; for example, vertebrate tail-bud mesoderm [2]). Further, comparative developmental genetic evidence supports the homology of the endomesoderm [4] and ectodermal neurogenesis [5] across eumetazoans. However, it remains controversial whether the cell layers in sponges are homologous to eumetazoan germ layers and whether sponges undergo gastrulation.
Sponges consist of four classes -Demospongiae, Calcarea, Homoscleromorpha and Hexactinellida [6,7] -of simple animals that share a common body organisation. As juveniles and adults, they are sessile, tri-layered animals that lack true muscles and nerves. Sponges possess an internal network of canals and ciliated choanocyte chambers lined with epithelial cells, primarily endopinacocytes and choanocytes, and are separated from the external environment by another epithelial layer, the exopinacoderm (Figure 1; hexactinellids are constructed as a syncytium). Sandwiched between these epithelial layers is the collagenous mesohyl, which is enriched with multiple types of amoebocytes, including the pluripotent stem cell type, the archeocyte ( Figure 1). Water currents created by ciliary beating in choanocyte chambers enable most of the physiological requirements of sponges, including feeding, respiration, excretion and reproduction. This juvenile/ adult body plan is the outcome of the dramatic reorganisation of the radially-symmetrical, bi-or trilayered larva at metamorphosis [8].
Despite the occurrence of gastrulation-like morphogenetic processes in sponges, it is still unclear how sponge development relates to eumetazoan gastrulation, which is defined as a morphogenetic event that leads to the establishment of germ layers. Some authors consider cell movements that generate multi-layered embryos during sponge embryogenesis gastrulation (for example, [12,16]), although some argue that gastrulation occurs during sponge metamorphosis when an adult internal layer develops by EMT and transdifferentiation of larval ciliated epithelial cells (for example, [17]). Yet it remains unclear whether sponge cell layers display the defining characteristic of eumetazoan gastrulation. This uncertainty led others to question the existence of germ layers, and hence gastrulation, in sponges (for example, [18]).
Here, we show by cell-labelling experiments in the demosponge A. queenslandica that the cell layers established during embryogenesis lack fixed identities and do not directly give rise to the cell layers of the juvenile. In addition, inner epithelial choanocytes in the juvenile can alter their fate by transdifferentiating into a range of cell types, including cells of the outer epithelium. Thus, Figure 1 Evolution of germ layers and gastrulation. Animal multicellularity, with the larval and juvenile/adult body plans of extant metazoans possessing multiple cell layers, is depicted above. The arrow represents metamorphosis from larval to juvenile/adult forms. Cell layers are coloured: orange, inner layer; green, outer layer; and blue, middle layer. In the case of cnidarians and bilaterians, these correspond to endoderm, ectoderm and mesoderm (bilaterians only), respectively. The phylogenetic relationship of these clades and the sister group to metazoans, the choanoflagellates, is shown below, along with the evolutionary origin of multicellularity. The origin of germ layers and gastrulation is debatable, occurring either before or after the divergence of sponges and eumetazoans.
sponge cell layers lack fate determination and stability. However, we find that expression of the sponge orthologue to the highly conserved eumetazoan endomesodermal marker GATA [4] is consistently restricted to the inner layer of larvae and juveniles, suggesting a deep shared ancestry between the sponge internal cell layer and the eumetazoan endomesoderm.

Results and discussion
Germ layers by definition exhibit restricted cell fate [2,3]. Thus, to address whether sponge embryonic and early juvenile cell layers are germ layers, we employed cell-labelling techniques in the demosponge A. queenslandica to examine the fate of larval epithelial and internal cells at metamorphosis, as well as the stability and fate of differentiated choanocytes in newly established juvenile tissues at the so-called 'rhagon' stage (staging described in Additional file 1: Figure S1).
The swimming A. queenslandica larva metamorphoses into a feeding juvenile with choanocyte chambers within three days of attaching its anterior end onto a natural substratum (crustose coralline algae) at 25°C. Within hours of settling, the three-layered, bullet-shaped larva transforms into an encrusting mat. A previous celllabelling study in A. queenslandica indicated that external larval cells transdifferentiated into internal choanocytes in juveniles [10]. However, it was not determined whether all cell types found on the larval exterior are transdifferentiating at metamorphosis, and whether the larval epithelial cells differentiate directly into choanocytes or indirectly via an intermediate cell type. In other sponges, the larval epithelial layer has been reported to shed entirely [19], be phagocytised by archeocytes [20][21][22], differentiate into choanocytes through a nonciliated amoebocyte intermediate [23][24][25], or directly differentiate into choanocytes without loss of cilia [26] (reviewed in [8]). Using the terminal deoxynucleotidyl transferase dUTP nick end-labelling (TUNEL) assay for detecting apoptotic DNA fragmentation, we demonstrate that a large proportion of A. queenslandica larval epithelial cells undergo programmed cell death shortly after the onset of metamorphosis [see Additional file 2: Figure  S2]. TUNEL-labelled nuclear fragments can be observed readily in the metamorphosing animal, often as phagocytised bodies within archeocytes [see Additional file 2: Figure S2].
To study the fate of non-apoptosing larval epithelial cells, the outer layer was labelled with the plasmamembrane dye CM-DiI. We assume that the cell lineage will be indicated by retention of this label in their plasma membrane, while the plasma membrane of the cells that phagocytised apoptotic fragments of larval epithelial origin will not be labelled. An overnight incubation of freeswimming A. queenslandica larvae in seawater containing 10 μM CM-DiI strongly labelled the plasma membrane of ciliated epithelial cell types, the columnar epithelial cell and the flask cell ( Figure 2A). During metamorphosis, these labelled cells internalised, resorbing their cilia and differentiating into archeocytes, which migrated throughout the animal [see Additional file 3: Figure S3]. Within two to three days, a number of juvenile cell types were labelled, including choanocytes and exopinacocytes ( Figure 2B, C), suggesting that these differentiated from pluripotent archeocytes, which were originally derived from the larval epithelium. Therefore, at metamorphosis ciliated epidermal cells of the larvae can either undergo apoptosis or transdifferentiate, via an archeocyte intermediate, into a range of juvenile cell types that are found in the inner, outer or middle layer of the juvenile.
Using the thymidine analog, 5-ethynyl-2′-deoxyuridine (EdU), to label the nuclei of proliferating cells, we found that only amoeboid cells localised to the inner cell mass, identified here as archeocytes [see Additional file 4: Figure S4], and no other cell types, continue to proliferate in the swimming larva ( Figure 2D). Taking advantage of this observation we followed the fate of these EdU-labelled cells during metamorphosis. We found that these archeocytes of the larval inner cell mass origin also gave rise to choanocytes and exopinacocytes during metamorphosis ( Figure 2E, F). Together, these observations indicate that a range of larval cell types can generate the diversity of cell types present in the juvenile and that there is no correspondence between the cell layers established during embryogenesis and those produced at metamorphosis.
Sponge metamorphosis has been regarded by some as gastrulation (summarised in [8,17]). As observed here and in previous studies, morphogenetic processes operational during sponge metamorphosis occur also during eumetazoan gastrulation (for example, EMT; reviewed in [8,27]). Therefore, we sought to determine if cell layers established at the end of A. queenslandica metamorphosis are developmentally determined. To test this, we labelled choanocytes that form internal choanocyte chambers and a small number of archeocytes surrounding choanocyte chambers by incubating juveniles with CM-DiI ( Figure 3A, B). Within one week, the majority of the labelled choanocytes had undergone EMT and dedifferentiated into archeocytes ( Figure 3C, D; Additional file 5: Figure S5), of which some had migrated away from the chambers and differentiated into other cell types including exopinacocytes ( Figure 3E) and sclerocytes ( Figure 3F). Thus, inner epithelial choanocytes are capable of transdifferentiating into outer epithelial cells via an archeocyte intermediate, demonstrating the lack of cell fate determination.
The lack of fate determination in cell layers in A. queenslandica is consistent with this demosponge lacking germ layers as classically defined. In light of this finding, genes shown to be instrumental in eumetazoan ectoderm or endomesoderm formation may not be expected to be present in demosponges. Indeed, of the genes previously implicated in the differentiation of the endomesoderm across Cnidaria and Bilateria, namely twist, snail, forkhead and GATA [4], only GATA was identified in the genome of the demosponge A. queenslandica [28] [see Additional file 6: Figure S6]. Also, amongst the 51 transcription factor-and signalling molecule-encoding genes shown to be 'endomesodermally' expressed in the cnidarian Nematostella vectensis [29], only nine orthologues are present in the A. queenslandica genome [see Additional file 7: Table S1].
A. queenslandica GATA transcripts are localised to a substantial fraction of internal cells of late-stage embryos, free-swimming larvae ( Figure 4B-D) and juveniles ( Figure 4F-I), but not detectable in outer layer epithelial cells at any developmental stage, similar to predominantly internal (that is, endomesodermal) expression of bilaterian GATA456 [30][31][32] and cnidarian GATA [4]. Thus, although this study argues against the existence of true germ layers in demosponges, the striking parallel in GATA expression patterns suggests a common ancestry between the sponge inner cell layer and the eumetazoan endomesoderm.
GATA genes are present in other early branching taxa, including placozoans, ctenophores and the other classes of sponges [33][34][35], although A. queenslandica is the only species for which there currently are published expression patterns. Understanding the evolution of germ layers will be assisted by the analysis of the evolution and expression of genes conserved in bilaterian mesoderm specification and formation, along with further analysis of developmental processes in these taxa. For example, recent analysis of the genome of the ctenophore Mnemiopsis leidyi failed to reveal many 'mesodermal' genes despite it appearing to have genuine endomesoderm [34], and it is known that cell movements from one embryonic layer to another occur during metamorphosis in some medusozoan cnidarians [36,37]. These approaches, along with resolving the exact order in which these early-branching metazoans diverged from the lineage leading to bilaterians (that is, whether sponges or ctenophores are the sister group to all other animals and whether sponges are monophyletic [34,38]), provide an opportunity to reconstruct the evolution of germ layers and gastrulation.

Conclusions
Comparison of demosponge and eumetazoan development and body plans minimally provides insights into their last common ancestor, which appears to have existed some 800 million years ago [39]. Although the possibility that their last common ancestor possessed germ layers and these were lost in the demosponge lineage cannot be excluded, the lack of developmental commitment of larval and juvenile cell types in A.
queenslandica and homology of internal layers across a demosponge and eumetazoans are consistent with eumetazoan germ layers and gastrulation evolving after the divergence of demosponge and eumetazoan lineages. Thus, we propose that ancestrally metazoans consisted of multiple cell layers whose fate was labile. The evolution of the gene regulatory network endowing progressive determination of cell layers resulted in the emergence of primary germ layers and mechanisms to Note the presence of nuclear fragments (fr) presumably resulting from phagocytosis that has occurred during de-differentiation [see Additional file 5: Figure S5]. Arrowheads in E and F show a labelled exopinacocyte and a sclerocyte, respectively, indicating these cell types can also be produced from de-differentiating choanocytes. Nuclei are stained with DAPI. Scale bars: 100 μm (A, C), 10 μm (B, D-F). DAPI, 4',6-diamidino-2-phenylindole. segregate these layers in eumetazoans (that is, gastrulation). The origin of the majority of genes that have conserved roles in eumetazoan gastrulation and germ layer determination evolved after the divergence of demosponge and eumetazoan lineages (for example, Fkh, Twist, Snail), giving credence to this supposition. The conserved expression of GATA in cells located inside the body of all metazoans, along with conserved differential expression of Wnt and TGF-β ligands along demosponge and eumetazoan body axes [40], suggests that these genes were essential in providing critical positional information to cells in the first multicellular metazoans. These, along with eumetazoan-specific genes, later became essential components of the specification and determination of eumetazoan germ layers.

Animal collection
Adult A. queenslandica were collected at the Heron Island Research Station, Queensland, Australia. Larvae were collected upon release from adults and metamorphosing individuals were obtained by allowing larvae to settle and metamorphose on a plastic petri dish as previously described [41].

Cell lineage tracing with CM-DiI
The lipophilic tracer CM-DiI (Molecular Probes, C7000, Mulgrave, Victoria, AUS) was used to label ciliated epithelial cells in larvae and choanocytes in the juveniles. Larvae or juveniles were incubated in seawater with 10 μM CM-DiI. To differentially label the plasma membrane of ciliated epithelial cells, larvae were incubated for 12 hours. Incubation for one hour was sufficient to differentially label the majority of choanocytes in juveniles. Little labelling occurred in internal cells in larvae or juveniles. The labelled specimens were rinsed in fresh seawater several times and then allowed to develop further. Specimens were fixed as described below at various stages of development. Nuclei were labelled with the fluorescent dye 4',6-diamidino-2-phenylindole (DAPI; 1:1,000, Molecular Probes), and the specimens were mounted in ProlongGold antifade reagent (Molecular Probes).

EdU pulse-and-chase experiments
Larvae or juveniles were incubated in seawater containing 200 μM of the thymidine analogue, EdU (Click-iT EdU AlexaFluor 488 cell proliferation kit, C10337, Molecular Probes), for 16 to 18 hours to label nuclei of proliferating larval archeocytes, and for three hours to label S-phase nuclei in the juveniles. Following washes in fresh seawater, the juveniles were immediately fixed. The larvae with labelled archeocytes were allowed to metamorphose into the juvenile stage, which were then fixed. Fixed specimens underwent immunohistochemistry as described below. Following the immunohistochemistry procedure, fluorescent labelling of incorporated EdU was conducted according to the manufacture's recommendations prior to DAPI labelling.

Gene isolation and whole-mount in situ hybridization
AqGATA gene fragments were amplified with gene specific primers, by using complimentary DNA from mixed developmental stages as a PCR template. Gene specific primers were as follows: F1, ATGGAGAAAGCAGACGACGCTA TGC; F2, CTACCAAGACAGTTTTGTGG; R1, CTACATC AATTGCTGTGGTTGCATGG and R2, ACTTTGACTT CTTTGACTCG. Primers F2 and R1 were used to generate a 776 base template for a riboprobe synthesis. Riboprobe synthesis and in situ hybridization were conducted as described previously [42]. An antisense digoxigenin-labeled riboprobe was hybridised at a final concentration of 1 ng/μl.

Transmission electron microscopy
Free-swimming larvae were processed for transmission electron microscopy (TEM) by using the high-pressure freezing (HPF) technique. Procedures followed the previously established HPF protocol for shark retinal tissues [43] with the following modifications: 20% BSA in artificial seawater was used as a cryoprotectant instead of the teleost ringer/0.7% agarose. In addition, 1% OsO 4 was used in the freeze substitution cocktail. Longitudinal ultrathin sections were prepared and analysed.

Additional files
Additional file 1: Figure S1. Stages of development during metamorphosis in A. queenslandica. A: Free-swimming larva stage; the (parenchymella) larva is egg-shaped and swims with the pigmented ring and associated bundles of long cilia directed posteriorly [10]. The anterior is down. In B-F, metamorphosing postlarvae are viewed from the top. Additional file 5: Figure S5. Epithelial choanocytes can de-differentiate into archeocytes in A. queenslandica. Choanocytes in chambers (ch) in juveniles were labelled with CM-DiI as in Figure 3A, B and descendants of the labelled cells in juveniles one day later are shown. Some choanocytes are seen to extend basal processes (ba) laterally, engulfing neighboring choanocytes (arrowheads in A, B). Some cells appear multinucleated (arrowheads in C), presumably as a result of phagocytic or fusion events. Abbreviations: ch choanocyte chamber; ci cilium. Scale bar: 10 μm.
Additional file 6: Figure S6. GATA sequence alignment and phylogenetic analyses. A: Partial GATA protein multiple sequence alignment showing the amino acid sites used for phylogenetic analyses. GATA genes encode zinc-finger transcription factors that bind to the (T/A (GATA) A/G) cis-regulatory motif [45] and are characterised by the presence of two zinc-finger domains (N-terminal 'N-finger' and C-terminal 'C-finger'). The alignment encompasses N-finger (underlined in red) and C-finger (underlined in blue) zinc-finger domains each encoding CXNC (boxed) X 17 CNXC (boxed). B: GATA maximum likelihood phylogeny rooted with GATA-type sequences. Amphimedon queenslandica sequence is highlighted in red. No clear GATA orthologues were identified outside Metazoa, and in the genome of the ctenophore Mnemiopsis leidyi [46]. C: Unrooted GATA maximum likelihood phylogeny. Sequence alignment and phylogenetic analyses were performed on the Geneious platform (v.5.1.7). Related sequences were retrieved via the protein BLAST search using the Amphimedon GATA-like sequence as queries, from GenBank at NCBI [47], Acropora digitifera genome (Version 1.1) portal at OIST [48], Mnemiopsis genome project portal at NIH [46] and Compagen at the University of Kiel (Oscarella sequences; [49]). Sequence IDs/accession numbers are shown with the name of each sequence in B and C. Peptide sequences were aligned with MUSCLE (v3.7) [50] (default settings), and ambiguous regions were manually removed. Phylogenetic trees were reconstructed using the maximum likelihood method implemented in the PhyML program [51]. The WAG substitution model [52] was selected assuming an estimated proportion of invariant sites and four gammadistributed rate categories to account for rate heterogeneity across sites. The gamma shape parameter was estimated directly from the data. Reliability for internal branches of maximum likelihood trees was assessed using the bootstrapping method (100 bootstrap replicates). Support values are shown at each node except when lower than 50%. The unit of the branch length is the number of substitutions per site. Additional file 7: Table S1. A list of the presence or absence of A. queenslandica homologues (orthologues and gene family members) to cnidarian Nematostella vectensis endomesodermal genes (listed in [29]) based on previously published data [28,[53][54][55][56][57][58] and sequence similarity searches in NCBI databases (via BLASTp).