The glycoproteins EDIL3 and MFGE8 regulate vesicle-mediated eggshell calcification in a new model for avian biomineralization

The avian eggshell is a critical physical barrier, which permits extra-uterine development of the embryo. Its formation involves the fastest known biomineralization process in vertebrates. The eggshell consists of proteins and proteoglycans that interact with the mineral phase to impart its specific microstructure and mechanical properties. In this study, we investigated the role of epidermal growth factor (EGF)-like repeats and discoidin-like domains 3 (EDIL3) and milk fat globule–EGF factor 8 (MFGE8), two glycoproteins that are consistently detected in eggshell proteomes. We verified their common evolutionary history and identified the timing of the duplication event giving rise to these two distinct proteins. Edil3/mfge8 chromosomal locations revealed a nested syntenous relationship with other genes (hapln1/hapln3 and vcan/acan) that are also involved in vertebrate calcification. EDIL3 and MFGE8 proteins possess EGF-like and coagulation factor 5/8 (F5/8C) domains, and their 3D structures predicted that they bind calcium and extracellular vesicles. In chicken, we confirmed the presence of EDIL3 and MFGE8 proteins in eggshell, uterine fluid, and uterus. We observed that only edil3 is overexpressed in tissues in which eggshell mineralization takes place and that this overexpression occurs only at the onset of shell calcification. We therefore propose a model in which EDIL3 and, to a lesser extent, MFGE8 proteins guide vesicles containing amorphous calcium carbonate to the mineralization site. This model was supported by the observation that extracellular vesicles accumulate in uterine fluid during eggshell calcification and that they contain high levels of calcium, carbon, and oxygen that correspond to calcium carbonate.

The avian eggshell is a physical barrier against mechanical and microbial stresses that allows extra-uterine development of the embryo in a closed chamber. The eggshell consists of 95% calcium carbonate (CaCO 3 , calcite), 3.5% organic matrix (OM 2 ; proteins, polysaccharides, and proteoglycans), and 1.5% water (1,2). This biomineral is composed of five layers: inner and outer eggshell membranes, two contiguous calcified layers (mammillary knobs and palisades), and the cuticle (3). Shell formation is extracellular and is the fastest biomineralization process in vertebrates (4). Egg formation occurs in the hen reproductive tract (1,5). In chickens, the forming egg reaches the red isthmus (RI) at 4.5 h post-ovulation (p.o.), where shell biomineralization is initiated on organic cores. The egg then enters the uterus (Ut), where it will remain for 18 -19 h during shell calcification. Calcium carbonate is initially deposited as an amorphous phase (ACC), which progressively transforms into calcite (6).
During its formation, the shell is bathed in a uterine fluid (UF) secreted by uterine cells that contains the organic and mineral precursors necessary for shell calcification (7,8). Matrix proteins stabilize ACC, promote crystal nucleation, select the calcite polymorph, and regulate the evolution of crystal size and morphology (4,7,9,10). These matrix-mineral interactions determine the orientation of calcite crystals, which results in the complex ultrastructure of the eggshell, its texture, and consequently its mechanical properties. Over the last couple of decades, numerous studies have defined the chicken eggshell matrix proteome (11)(12)(13)(14)(15)(16)(17)(18). Currently, the entire dataset of reported proteins contains around 900 nonredundant proteins for the chicken eggshell and more than 600 for the UF (19,20). Among them, EDIL3 (EGF)-like repeats and discoidin-like domains 3) and MFGE8 (milk fat globule-EGF factor 8) are reported in the majority of transcriptomics and proteomics studies associated with chicken eggshell biomineralization (11, 13-18, 21, 22). These two proteins were also identified in other avian eggshell proteomes (quail, turkey, and guinea fowl) (23)(24)(25).
MFGE8 and EDIL3 are two secreted glycoproteins that have been well-described in mammals and exhibit sequence similarity to each other. In avian eggshell, it was suggested that both EDIL3 and MFGE8 could bind Ca 2ϩ , due to an EGF-like calcium-binding domain, and could favor crystal nucleation or change crystal morphology by selective interaction with calcium ion-rich crystallographic faces (19,21).
Nevertheless, the function of EDIL3 and MFGE8 is poorly understood. In this study, we have used bioinformatics tools to decipher the common evolutionary history of edil3 and mfge8 genes. We also explored the role of EDIL3 and MFGE8 in chicken eggshell mineralization by predicting their molecular features, quantifying their gene expression, and measuring protein levels in tissues involved in eggshell formation. Our study suggests that EDIL3 and MFGE8 would bind to vesicles and calcium carbonate to guide vesicular transport providing mineral cargo during avian eggshell biomineralization.

Evolutionary history of EDIL3 and MFGE8: synteny and phylogenetic analysis
EDIL3 and MFGE8 are two related proteins that share 54% identity and 69% similarity in Gallus gallus (EDIL3, XP_424906.3; MFGE8, NP_001264039.1; Table S1). They are encoded by two distinct genes, edil3 and mfge8, which are localized on chromosome Z and chromosome 10, respectively. We compared their gene loci in different vertebrate classes (fishes, amphibians, reptiles, birds, and mammals; Fig. 1). Edil3 and mfge8 were observed to be nested in syntenous clusters containing hapln1 (hyaluronan and proteoglycan link protein 1) and vcan (versican) adjacent to edil3, and hapln3 (hyaluronan and proteoglycan link protein 3) and acan (aggrecan) adjacent to mfge8 (Fig. 1). HAPLN1 and HAPLN3 belong to hyaluronan and proteoglycan link protein family, and VCAN and ACAN belong to the lectican family. In chicken, HAPLN1 is related to HAPLN3 with 48% identity/66% similarity for conservative substitutions, whereas VCAN and ACAN displayed 18% identity/27% similarity (Table S1). The direction of mfge8 gene cluster transcription is unchanged from Chondrichthyes (Callorhinchus milii) to mammalia (Homo sapiens) (Fig. 1). This observation is also true for the edil3 cluster during vertebrate evolution (Fig. 1). Therefore, the edil3 and mfge8 clusters are highly conserved in Gnathostoma. In birds, the edil3 gene cluster is consistently observed on chromosome Z, and the mfge8 cluster is found on a nonsexual chromosome (Table S2). These results confirm that the edil3 and mfge8 gene loci contain paralogous genes resulting from a duplication event.
To determine the timing of divergence between these two genes, we performed phylogenetic tree analysis of the protein sequences to investigate the evolutionary history of EDIL3 and MFGE8 in metazoans (Fig. 2). We first investigated the presence of the two proteins in eukaryotes. Blast analysis revealed that EDIL3 and/or MFGE8 were absent from protozoans to sponges. Consequently, the phylogenetic reconstruction was performed with EDIL3/MFGE8 protein sequences from eumetazoan species (Fig. 2). The NCBI database contains a large number of EDIL3/MFGE8 orthologs. To perform this analysis, we loaded 34 EDIL3/MFGE8 sequences from 19 different species representing different taxa. These sequences belong to nonvertebrates (one cnidarian, one platyhelminthe, two molluscs, one arthropod, one echinoderm, and one cephalochordate) and vertebrates (two fishes, one amphibian, three reptiles, three birds, and three mammals; Table S3). MFGE8 sequences were often annotated as lactadherin in various databases (Table S3). For consistency during our analysis, all lactadherin sequences were renamed MFGE8 (Fig. 2). Sequences were aligned using Clustal Omega multiple alignment. A phylogenetic tree was constructed using Maximum Likelihood and Bayesian inference methods, which both conducted to similar tree topology, and Maximum Likelihood result was displayed (Fig. 2). According to the topology of the tree, with coral Acropora digitifera EDIL3 protein as a rooted sequence, two distinct groups emerged within the phylogenetic tree. The first one contained nonvertebrate EDIL3/MFGE8 sequences, and the second one contained vertebrate EDIL3/MFGE8 sequences. The latter group had a cephalochordate MFGE8 sequence as a root and is then divided into two clusters: vertebrate EDIL3 and vertebrate MFGE8 sequences, respectively. Inside both clusters, we noticed that species were also grouped according to their class (actinopterygii, amphibian, reptiles, aves, and mammals).
In Fig. 2, the number of substitutions by site indicated that EDIL3 had undergone fewer changes during vertebrate evolution compared with MFGE8. Indeed, MFGE8 branches were longer than EDIL3 branches, and the EDIL3 cluster bootstrap value (node value, 98) was higher than the value for the MFGE8 cluster (node value, 56). This result indicated that vertebrate sequences of MFGE8 were more divergent from each other than within the EDIL3 sequences. In nonvertebrates, EDIL3 and MFGE8 sequences did not form two distinct clusters. Moreover, we did not consistently observe both EDIL3 and MFGE8 protein sequences; in several nonvertebrates species, only one sequence is present. These are mostly annotated as lactadherin-like (Table S3). The presence of both genes in the same nonvertebrate species, and their proximity in the tree, suggested that another duplication event affecting the ancestral edil3/mfge8-like gene appeared in the nonvertebrate lineages, as observed for the coral A. digitifera, the cephalochordate Branchiostoma belcheri and the sea urchin Strongylocentrotus purpuratus. We also observed during the course of vertebrate evolution a distinction between edil3 and mfge8 emerging around 480 million years ago (MYA), which led to further diversification/divergence of their sequences and gave rise to vertebrate edil3 and mfge8. This event occurred after the split of nonvertebrate/vertebrate lineages and before the split between fishes/amphibian phyla. Therefore, the edil3/mfge8 divergence appeared between 684 and 473 million years ago.
A comparison of protein domains was performed throughout eumetazoan evolution (Fig. 2). In nonvertebrates, only EDIL3 of the coral A. digitifera possesses EGF-like domains, although this domain was present in all EDIL3/MFGE8 vertebrate species. The most primitive vertebrate of the study (C. milii) possessed three EGF-like domains in MFGE8 but only two in its EDIL3. With the notable exception of fishes, all EDIL3 vertebrate sequences contain three EGF-like domains. Three EGF-like domains are also present in nonmammalian MFGE8 sequences, whereas two and one are present in nonprimate EDIL3/MFGE8 in avian shell calcification mammals and primates, respectively. A Ca 2ϩ -binding motif is also present in A. digitifera EDIL3 and in all other sequences except in fishes EDIL3, mammalian MFGE8, and nonvertebrates. An RGD motif (integrin (ITG)-binding) was also detected in EGF-like domains of all vertebrate MFGE8s and EDIL3s. In nonvertebrates, this motif is observed in F5/8C domains, when present. This latter domain was observed in all vertebrate and nonvertebrate sequences. A double RGD motif was also detected in three nonvertebrate sequences and in G. gallus EDIL3. However, the vertebrate RGD sequences are

EDIL3/MFGE8 in avian shell calcification
found in the EGF-like domains, whereas the invertebrate RGDs are located within the F5/8C regions.

Edil3 and mfge8 expression in G. gallus tissues at various stages of shell mineralization
We investigated the level of expression of both genes in oviduct tissues involved in egg white and eggshell formation (Fig.   3A). These are the magnum (Ma), where egg white proteins are synthesized, and the white isthmus (WI), where the shell membranes are formed. In addition, the red isthmus (RI) and uterus (Ut) are involved in the onset and development of shell mineralization, respectively. Gene expression in these oviduct segments was compared with samples from mid-shaft tibial bone (B) (mineralized tissue), duodenum (D) (high calcium intake), LG substitution model) was performed using Clustal Omega, and the phylogenetic tree was reconstructed using Maximum Likelihood (in MEGA7) and Bayesian inference (in BEAST version 1.10.4) methods. The topology of the tree is that resulting from Maximum Likelihood. Node values correspond to the percentage of Maximum Likelihood bootstrap values/Bayesian posterior probabilities, respectively. Only values above 50 are indicated. The divergence times were analyzed on Timetree. The schematic representation of EDIL3 and MFGE8 protein domains were acquired with HomoloGene of NCBI and several databases (ExPASy-PROSITE, Ensembl, and SMART). RGD, integrin-binding site; Ca 2ϩ , calcium-binding site. NCBI accession numbers for each protein are listed in Table S3.

EDIL3/MFGE8 in avian shell calcification
kidney (K) (involved in ion exchange), and finally liver (L) as an important tissue of general metabolism. Levels of expression were measured in tissues harvested during the onset of the rapid growth phase of shell mineralization (10 h p.o., except for B collected at 18 h p.o. when shell mineralization and bone demineralization are completed).
Edil3 expression was low in D, L, Ma, B, and K and not significantly different from background levels (Fig. 3A). Edil3 was significantly expressed in tissues involved in shell formation in contrast to the five other tissues. They were, however, not significantly different from each other. RI/B and Ut/B expression ratios were 58.6 and 39.2, respectively, and 846.9 and 566.9, respectively, for RI/K and Ut/K ratios. Mfge8 was expressed in all tissues (Fig. 3A). Liver exhibited the highest level of mfge8 expression followed by RI. Mfge8 expression levels were significantly different in these two tissues, which were both significantly different from the six other tissues. Ratios of RI expression were 73, 18.3, 4.1, 2.5, and 2.6 with Ma, D, B, K, and WI, respectively.
Levels of edil3 and mfge8 expression in oviduct regions that are active during eggshell mineralization (Ut and RI) were measured at five key stages of shell calcification. Edil3 was expressed at all stages in Ut and RI and was significantly higher in Ut for stages 6 h and 7 h p.o. compared with 16 h p.o. (Fig. 3, B and C), when ACC is massively deposited with the first aggregates of calcite forming around mammillary knobs. Edil3 expression compared with 16-h p.o. exhibited a fold change of 5.6 and 4.9 at 6 h and 7 h p.o., respectively. In contrast, from 5 h to 16 h p.o., no significant difference was observed for edil3 in RI and for mfge8 in Ut and RI (Fig. 3, B and C).

EDIL3 and MFGE8 protein levels during eggshell formation
The presence of EDIL3 and MFGE8 proteins was investigated in soluble eggshell matrix (SEM), UF, Ut, and D (which is involved in ion transfer without mineralization). Using anti-EDIL3 antibodies, a single immunoband was observed around 60 kDa (expected size 54 kDa) in SEM, Ut, and to a lesser extent in UF (Fig. 3D). No immunoreactive band was observed in D. MFGE8 exhibited an intense immunoreactive band in all tissue samples, including D (Fig. 3D) at 75 kDa (expected 53 kDa). The higher than predicted molecular mass observed for MFGE8 could be due to post-translational modification such as glycosylation, which has been reported for this protein (26,27). These results are in accordance with expression measurements, and they confirmed that EDIL3 is specific to the oviduct tissue involved in shell mineralization, whereas MFGE8 is present in a variety of tissues or organs, including those responsible for eggshell calcification (Fig. 3).

Characterization of G. gallus EDIL3 and MFGE8 functional domains
We explored gene and protein sequences of EDIL3 and MFGE8 using bioinformatics tools. G. gallus edil3 is made of 11 exons and is located on sex chromosome Z (Fig. 4). Its transcription leads to two different transcripts of 5544 and 1973 nucleotides coding for 480 and 470 amino acid proteins, respectively (NCBI). Differences in transcript size are the consequence of a 30-nucleotide deletion in exon 3 and alternative splicing of exon 11. This latter alternative splicing does not affect the translated protein because it modifies the 3Ј-UTR region in the mRNA. Stop codons are located at positions 1657-1659 and 1687-1689 in the short and long mRNA variants, respectively. The 30-nucleotide in-frame deletion of exon 3 resulted in a loss of 10 amino acids between two EGF-like domains. The G. gallus mfge8 gene is located on chromosome 10 and is made of 10 exons (Fig. 4). As is the case for edil3, mfge8 also exhibited two transcripts of 2588 and 2570 bases, which encode proteins of 474 and 468 amino acids, respectively (NCBI). The shorter transcript is the result of an 18-nucleotide deletion at the onset of the last exon, which induces a loss of six amino acids between two F5/8C domains of the corresponding protein. The remaining sequence is not affected by this deletion.
EDIL3 and MFGE8 G. gallus proteins were aligned with MEGA7 software, and protein domains were investigated using PROSITE (Fig. 5). Both proteins exhibited three EGF-like domains at the N terminus (

EDIL3/MFGE8 in avian shell calcification
(Asn-109, Asn-111, Asp-126, and Tyr-131) were highly conserved when compared with the EGF-like Ca 2ϩ -binding consensus sequences of the human protein (Asp-1, Asp-3, Asp-18, and Tyr-23; Fig. 5 and Fig. S1A). However both Asps of the human coagulation factors IX/X Ca 2ϩ -binding motif were replaced by Asn (residue with a similar polar property to Asp) in each G. gallus protein. Cysteines were strictly conserved between both proteins at the Ca 2ϩ -binding site.
We performed computational predictions of secondary structure for chicken EDIL3 and MFGE8 using PSIPRED version 3.3 and the I-TASSER program server. Twenty two ␤-sheets and two ␣-helices were predicted for EDIL3, and 25 ␤-sheets and two ␣-helices were predicted for MFGE8 proteins, with good correspondence between the 22 ␤-sheets of EDIL3 and those in MFGE8. We also observed that EGF-like domains are made up of two ␤-sheets in each protein. The first F5/8C contains 11 ␤-sheets in MFGE8 and nine in EDIL3, whereas the second is similar in both proteins with seven ␤-sheets.
The 3D structures of EDIL3 and MFGE8 were also predicted with the I-TASSER program server (molecular homology modeling), and the spatial arrangement of amino acids involved in each binding function (ITG, Ca 2ϩ , and PS) was highlighted on the 3D structure models using PyMOL Molecular Graphics system (Fig. 6, A and C). A close spatial proximity among residues was observed in accordance with their ability to interact with their substrates, and consequently, we predicted that these domains are functional. The conformation of the Ca 2ϩ -binding domain was composed, as expected, of three disulfide bridges necessary to maintain the protein conformation for interaction with Ca 2ϩ (Fig. 6, A and C). The spatial arrangement of residues in the second F5/8C domains associated with the PS-binding motif in G. gallus is highlighted (Fig. 6, A and C). This motif involves seven and eight amino acids for EDIL3 and MFGE8, respectively. Residues are located in three different loops as observed for the mammalian sequences (Fig. 6, A and C, green balls). Only Gln-390 was observed in a ␤-sheet in the MFGE8 F5/8C 2 domain. Both EDIL3 and MFGE8 RGD motifs involved in ITG binding of the second EGF-like domains were localized to a loop between two ␤-sheets (Fig. 6, orange balls). Structural modeling also indicates that the major difference between EDIL3 and MFGE8 is associated with the first EGF-like domain (Fig. 6), for which an additional specific RGD motif was observed in EDIL3. This additional RGD motif is not located in a loop between ␤-sheets.

Discussion
EDIL3 and MFGE8 are two proteins widely detected in proteomic studies associated with chicken eggshell biomineralization (11,(13)(14)(15)(16)(17)(18)21). In chicken these two proteins exhibited a high abundance in partially mineralized eggshell at the initiation of biomineralization, and consequently, they are thought to be potential major actors in the calcification process (21).
MFGE8 and EDIL3 are homologously-secreted glycoproteins that have been well-described in mammals. The mfge8 transcript was first identified in mouse mammary epithelial cells (29). MFGE8 is known to participate in a wide variety of cellular interactions such as macrophage and apoptotic cell bridging, adhesion between sperm and the zona pellucida of the egg, or exosome function (30 -34). EDIL3 was first identified in mouse as an extracellular matrix protein involved in embryonic vascular development and demonstrating an angiogenic role (35)(36)(37). In mammals, EDIL3 is physiologically expressed in various tissues such as brain, intestine, heart, and kidney (38). EDIL3 was also detected during cancer development and is expressed by primary tumors from breast and colon cancers (39). In breast cancer, EDIL3 was identified on extracellular vesicles (EV) promoting cellular invasion and was suggested to be a potential biomarker for monitoring this metastatic cancer (40). In developing chicken embryos, experiments have been performed on chorioallantoic membrane (CAM) using mammalian EDIL3 (DEL1). DEL1 was found to be involved in embryonic vascular remodeling, in mediating the breakdown of existing vascular blood, and inducing restructuring of the CAM vasculature (35). In contrast, a potent pro-angiogenic activity was reported for DEL1 in chicken CAM, which was mediated by integrin (␣v␤3) activation via the RGD motif (36).
Our phylogenetic analysis demonstrated that vertebrate edil3 and mfge8 are derived from a duplication event of an ancestral common gene. The divergence point between these two paralogs was estimated to be about 480 MYA. Gene duplication processes are crucial to generate genes with novel or altered functions (41). The duplication events giving rise to edil3 and mfge8 in early vertebrates emerged prior to the appearance of the amniotic egg at the split between amphibians (uncalcified egg) and reptiles (calcified egg, amniota) that

EDIL3/MFGE8 in avian shell calcification
occurred around 350 MYA (42,43). According to our study, the function of EDIL3 and MFGE8 was recruited to diverse biological processes, including eggshell calcification.
In nonprimate mammals, two homologous domains of epidermal growth factor from Drosophila NOTCH (EGF-like) at the N terminus and one tandem of discoidin-like domains similar to those of coagulation factors V and VIII (F5/8C) at the C terminus were described in MFGE8 (29,44). Our evolutionary study confirmed the presence of these domains in mammals and also explored additional eumetazoan sequences used in this study. The F5/8C domains appear to be present in all species sequences indicating an origin of this domain in the common ancestor of eumetazoans. The number of F5/8C domains is variable in nonvertebrates (from two to four), whereas it is constant in vertebrates, in which two F5/8C were reported. The PS-binding site of MFGE8 orthologs was previously described in cow (Bos taurus) and mouse (M. musculus) (28,(45)(46)(47)(48). The following residues (Lys-24, Trp-26, Phe-31, Lys-45, Gln-76, Arg-79, Phe-81, and Arg-146) are major residues contributing to the interaction between the F5/8C 2 domain of MFGE8 and PS in the mouse (28). Several studies reported the mechanism of mammalian MFGE8 F5/8C binding to PS (47,49). PS binding allows MFGE8 to complex with vesicles or cell membranes through their second F5/8C domain to perform physiological functions such as cellular interactions and exosome transport (30,31,33). Thus, mammalian MFGE8 binds to PS-containing vesicles in a two-step mechanism that explains the strong enrichment of MFGE8 in exosomes from dendritic cells (30,47). We identified this motif in the G. gallus EDIL3/MFGE8 sequences suggesting that chicken EDIL3 and MFGE8 proteins can also bind to membrane phospholipids.
EGF-like domains are present in all vertebrates but absent in nonvertebrates with the notable exception of EDIL3 of the Cnidaria A. digitifera, which is the most primitive eumetazoan species used in this study. In vertebrates, three EGF-like domains were present, except for EDIL3 in fishes (two domains) and MFGE8 in mammals (one and two domains in primate and nonprimates, respectively). This result confirms previous findings on mammalian MFGE8 (27,50). The highest-conserved EGF-like domain in vertebrate species includes an RGD motif (arginine, glycine, and aspartic acid) in both MFGE8 and EDIL3 proteins. This motif displays a strong affinity for ␣v␤3/␣v␤5 ITGs, which are expressed at the cell surface and widely present in vertebrates, underlying its major role in cellular interactions (30,31,33,34). It is present in the second and third EGF-like domains in vertebrates. In nonvertebrates, the RGD site is found in the F5/8C domain of protostomian species (arthropods, molluscs, and trematodes) and is completely absent in others (Fig. 5). These results suggest the independent acquisition of the RGD motif in protostomians and in vertebrates. We also predict a Ca 2ϩ -binding motif in the third EGF-like domain of EDIL3 and MFGE8 in the majority of the vertebrate sequences analyzed. However, this motif was not observed in mammalian MFGE8 or in EDIL3 of fishes selected for this study. The Ca 2ϩ -binding motif was therefore lost in MFGE8 at the split of oviparous/viviparous vertebrates. Interestingly, MFGE8 was reported as the major component of the milk fat globule membrane, which is a calcium-rich milieu (51). In nonvertebrate species, Ca 2ϩ -binding sites were only found in the EDIL3 sequence of the staghorn coral (A. digitifera), which precipitates a calcium carbonate exoskeleton. However, we did not identify Ca 2ϩ -binding motifs in protein sequences of the purple sea urchin (S. purpuratus) or the eastern oyster (Crassostrea virginica), which also mineralize a CaCO 3 skeleton. The cephalochordate B. belcheri, which is a "fish-like" species without a calcareous skeleton, did not exhibit this motif.
The question of the role(s) of these two proteins in avian eggshell formation is the main focus of this study. EDIL3 and MFGE8 were initially proposed to be important molecular actors in chicken eggshell mineralization (21). They were quantified at the four key steps of shell mineralization and classified according to their abundance (21). Among the 216 eggshell proteins quantified in this study, EDIL3 abundance was among the top 10 proteins at the two stages corresponding to the transformation of ACC into calcite crystals. The abundance of MFGE8 was intermediate in the shell, whatever the stage of shell mineralization (21). A comparative study revealed that MFGE8 is overabundant in the proteome of strong eggshells compared with weaker shells (15). In addition to their detection in numerous investigations of the chicken eggshell proteome, both proteins were identified in other avian eggshell proteomes (quail, turkey, and guinea fowl) (23)(24)(25). The experiments performed in this study support our hypothesis that EDIL3 plays a key role during eggshell formation. Edil3 was overexpressed in the oviduct regions where eggshell formation takes place (WI, RI, and Ut), and EDIL3 protein was present in SEM, UF, and Ut. We also compared edil3 expression levels with those of other uterine genes during eggshell calcification. We have previously observed by RNA-Seq that the edil3 counting per million (CPM) value was about three times higher than the median CPM value of genes coding for eggshell matrix proteins (52,53). The median emPAI of proteins quantified in the shell proteome is 5.92, whereas the maximum emPAI was obtained for lysozyme with a value of 13537. The second most abundant protein was Ovocleidin-116 (OC-116, MEPE) with an emPAI value of 7168. EDIL3 exhibited a value of 3523 and was the sixth most abundant eggshell protein. MFGE8 was the 34th most abundant protein with an emPAI value of 37.5. The role of MFGE8 in chicken eggshell biomineralization is not so clear. Indeed, mfge8 was significantly overexpressed in RI where shell mineralization is initiated, but its expression is not specific to the oviduct segment where shell mineralization occurs; moreover, it is ubiquitous in all tested tissues, with the highest levels in liver. Nevertheless, its levels in uterus determined by RNA- Figure 5. Domain prediction for G. gallus EDIL3 and MFGE8 protein sequences. GenBank TM accession numbers used are EDIL3 XP424906.3 and MFGE8 NP_001264039.1. Alignment was performed using Clustal Omega method in MEGA7 and then annotated with BioEdit. Red-and blue-colored residues indicate identity and similarity, respectively. Blue and yellow backgrounds correspond to EGF-like and F5/8C domains, respectively. Red vertical arrows (filled for cysteines) and green circles correspond to amino acid residues involved in Ca 2ϩ binding and PS binding, respectively. The letters below arrows and circles indicate the consensus pattern for Ca 2ϩ -binding and PS-binding residues of human coagulation factors IX/X and mouse MFGE8, respectively. Indicated positions correspond to the residue numbering for the full-length sequences of G. gallus EDIL3 and MFGE8.

EDIL3/MFGE8 in avian shell calcification
Seq were about two times higher than the median CPM value for genes encoding matrix proteins (52,53).
In mammals, MFGE8 was reported to be involved in spermatozoid-egg zona pellucida-binding (54). In bird eggs, the vitelline membranes that surround the yolk contain zona pellucida proteins that originate from the liver (55). Therefore, MFGE8 could play a role in spermatozoid adhesion. Western blotting confirmed the presence of MFGE8 protein in shell, uterine fluid, and uterus but also in duodenum. In birds, the duodenum is not involved in shell calcification but plays a key role in the uptake of dietary calcium to provide the ions necessary for bone and eggshell formation (56).
Our exploration of edil3 and mfge8 chromosomal arrangements throughout vertebrate evolution highlighted a strongly conserved synteny for both genes (Fig. 1). In avian species, we observed that edil3 and mfge8 are localized on chromosome Z and on a nonsexual chromosome, respectively. Interestingly, quantitative trait loci affecting chicken eggshell strength are present on the Z chromosome (57,58). Moreover, we observed that the adjacent genes at the edil3 and mfge8 loci are hapln1/ vcan and hapln3/acan, respectively. HAPLN3 is one of the most abundant proteins in the chicken eggshell, and HAPLN1 was identified in the eggshell membranes of fertilized eggs (21,59). Proteoglycans play an important role in eggshell biomineralization; for example, the dermatan sulfate proteoglycan ovocleidin-116 (MEPE) is abundant in eggshells (21,(60)(61)(62)(63). Proteoglycan macromolecules combine a protein core with a negatively charged complex polysaccharide and strongly interact with calcium (64). In mammals, HAPLN1, ACAN, and VCAN are cartilage matrix proteins, and they participate in multimolecular aggregates (65)(66)(67).
Numerous studies on mammals revealed the participation of vesicles in bone and cartilage mineralization (68 -73). A noncrystalline calcium phosphate phase was shown within intracellular vesicles of bone-lining cells (74). Membrane-bound mineral particles have been identified in osteoblasts (75), and numerous vesicles containing calcium phosphate are also present in the circulatory system or in the extracellular space adjacent to osteoblasts during embryonic chicken long-bone development (76). Additionally, MFGE8 is an abundant protein in mineralizing vesicles from chicken embryonic femurs and was also present with EDIL3 in mammalian cartilage vesicles (77)(78)(79). Furthermore, HAPLN1, ACAN, and VCAN were also identified in these vesicles (77,78,80). Recently, an RNA-Seq study reported that several vesicular genes are highly expressed in hen uterus (52). For example, pdcd6ip and sdcbp encode proteins (PDCD6IP and Syntenin-1) found in matrix vesicles isolated from bone or cartilage (79). In addition, EM studies of hen utero-vaginal junction detected a large number of vesicles released into the lumen (81). Spherical voids can be visualized by scanning EM in the calcified eggshell (82,83), which we hypothesize are the remnants of mineralization-associated matrix vesicles. Spherical structures (Ϸ300 nm) are observed within the organic matrix of decalcified eggshell (82); they are observed throughout the entire palisade layer and upper mammillary knobs, and they are immunoreactive for the eggshell proteoglycan ovocleidin-116 (84 -87).
We hypothesize that extracellular vesicles are involved in eggshell formation. Chicken eggshell biomineralization requires ACC to supply ions for rapid mineralization (88,89). Indeed, ACC was reported as an important highly-reactive translational phase for the formation of sea urchin spines or molluscan shells (89 -91). In chicken eggshell, ACC is first massively deposited on the entire surface of the shell membranes, and then it accumulates at specific nucleation sites to supply ions to form calcite (6). ACC is highly unstable under physiological conditions, and consequently, the vesicles observed in eggshell could stabilize and transport ACC during mineralization. In vitro studies have shown that liposome vesicles can stabilize ACC and prevent its premature crystallization (92). Indeed, uterine fluid is a supersaturated solution (93); therefore, transport of ACC via vesicles through the uterine fluid would avoid ectopic calcification, as proposed for bone mineralization by calcium phosphate (75). We also reported in this study that G. gallus EDIL3 and MFGE8 have Ca 2ϩ -binding and PS-binding sites. PS binding would allow their attachment to vesicle membranes, and the Ca 2ϩ -binding domain could interact with the mineral phase. We therefore propose a model in which EDIL3 and MFGE8 would guide vesicles containing ACC cargo to the site of mineralization (Fig. 7). This hypothesis is supported by the description in dendritic cell exosomes of MFGE8 targeting effector cells ITG as a guiding protein (30). The presence of vesicles in CaCO 3 mineralization has been described in sea urchin embryos (94). In this study, EDIL3 and MFGE8 were observed in Ut and UF and would be present in vesicles originating from uterine epithelial cells, transiting through the UF and targeting Ca 2ϩ to the mineralization site to deliver the mineral precursor (Fig. 7). We used transmission electronic microscopy (TEM) to analyze uterine fluid, and we confirmed the presence of EV in the 100 -300-nm size range in this milieu (Fig. 8). We also performed energy-dispersive X-ray spectroscopy (EDS) on vesicles to detect elements present in vesicles (Fig. 9). Several mapping measurements were obtained across a vesicle (Fig. 9B) or mapped point by point at different sites where vesicles were present or absent (Fig. 9C). The spectra showed that calcium, carbon, and oxygen were present in vesicles in agreement with calcium carbonate. However, phosphorus and chlorine, representing alternative counterions for calcium (phosphate and chloride), were not detected at these levels. This result is robust evidence for calcium carbonate inside UF vesicles and highlights their contribution to the transport of calcium necessary for calcification.  EDIL3 (XP_424906.3). B, schematic representation of G. gallus EDIL3 and MFGE8 protein domains, including their binding sites. C, tertiary structure prediction of EGF-like and F5/8C regions of G. gallus MFGE8 (NP_001264039.1). The tertiary structure prediction models were generated by EGF-like and F5/8C domain regions splitting using the I-TASSER template-based modeling. Following prediction modeling, the generated tertiary structures were edited using PyMOL. Orange balls indicate RGD residues of the ITG-binding sites; red balls illustrate residues of the Ca 2ϩ -binding sites, and green balls display PS-binding sites. The position of the residues is indicated according to their location in the entire protein. Elements were from Servier Medical Art, licensed under a Creative Commons Attribution 3.0 Unported License.

EDIL3/MFGE8 in avian shell calcification
Moreover, we confirmed that both proteins are incorporated into the shell matrix during mineralization. Additionally, our results highlighted edil3 and mfge8 expression throughout eggshell formation and notably at the early stages for edil3 when ACC deposition is massive (6). This scenario would take place during the entire process of shell formation, as ACC is required at the mineralization front (6).
We surveyed previous chicken eggshell and uterine fluid proteomic studies (11,15,19,21) to identify additional proteins potentially involved in our proposed vesicular transport model (Table 1). Annexins (ANX) should be involved in vesicular transport as shown in vitro for cartilage or bone (95)(96)(97). Indeed, ANXs can act as a Ca 2ϩ channel in vesicle membranes (Fig. 7) (98 -100). ANXA1, ANXA2, ANXA5, and ANXA8 are detected in the chicken eggshell, and their encoding genes are highly expressed in uterine cells (11,15,21,52). ANXA2 is also identified in other bird eggshell proteomes such as duck, quail, turkey, and guinea fowl, suggesting a common role in shell biomineralization (25, 101). Eggshell CaCO 3 requires bicarbonate ions in addition to calcium to form ACC. This role of the HCO 3 Ϫ supplier inside the vesicles could be ensured by carbonic anhydrases (CAs), which transform soluble CO 2 into bicarbonate ions (Fig. 7). Indeed, the UF has a high partial pressure of CO 2 (100 mm Hg), which is elevated 2-fold compared with plasma (93,102). Among the members of the carbonic anhydrase family, CA4 is present in the eggshell proteome of chicken, duck, quail, turkey, and guinea fowl (25, 101). CA4 is a glycosylphosphatidylinositol-anchored protein that is attached to plasma membrane, and consequently, we predict that it is present in the extracellular vesicles. Ovocleidin-17 (OC-17), ovalbumin (OVAL), and lysozyme (LYZ) are three chicken eggshell proteins described as ACC stabilizers (103)(104)(105). Finally, we used the Vesiclepedia database (106) to explore vesicular markers among chicken shell proteins already identified (Table 1).
In this study, we used a bioinformatics approach to delineate the evolutionary history of edil3/mfge8 in Eumetazoa, and we suggest that EDIL3 and MFGE8 proteins in vertebrates are derived from a duplication event, as supported by observed synteny in these and adjacent genes. Our gene expression and immunochemistry results allow us to propose a coherent model for EDIL3 and MFGE8 function in vesicles that deliver stabilized ACC for shell mineralization. This scenario involves the vesicular transport of both Ca 2ϩ and HCO 3 Ϫ in the uterine fluid and predicts the involvement of additional vesicular genes, such as annexins. We confirmed the presence of extracellular vesicles in the uterine fluid and detected the elements of calcium, carbon, and oxygen (calcium carbonate) within them.

Ethical statement animal handling and housing
The present experiments, including animal handling protocols, were performed in accordance with European Communities Council Directives concerning the practice for the care and

Uterine fluid, egg, and tissue collections
Thirty six mature brown laying hens (ISA-Hendrix, 40 and 90 weeks old) were placed in individual furnished cages with automatic devices for recording of the oviposition times. Animals were fed ad libitum (layer mash) using a commercial feed for layers. Laying hens were subjected to a cycle of 14 h of light/10 h of darkness. Following collection and opening of eggs, the eggshells were thoroughly washed with water, air-dried, and stored at Ϫ20°C.
Uterine fluids were collected as described previously (7). Briefly, egg expulsion was induced by an intravenous injection of 50 g of Dinolytic prostaglandin-F2␣ (Zoetis, Malakoff, France) at 7-9 and 16 h p.o., which correspond to the initiation and active phase of shell mineralization. Immediately after egg expulsion, UFs were collected by gravity in a tube placed at the entrance of the everted vagina. UFs were directly frozen in liquid nitrogen in the form of beads or diluted into PBS. UF samples were stored at Ϫ20°C until use. An additional UF aliquot was also diluted 5ϫ for SDS-PAGE into Laemmli sample buffer, for a final composition of 0.0415 M Tris-HCl, pH 6.8, 1% SDS, 10% glycerol, 1.66% ␤-mercaptoethanol, and 0.01% bromphenol blue.
Tissues were collected from 30 40-week-old and six 90-weekold laying hens. Animals were euthanized with Dolethal (Vetoquinol, Magny-vernois, France) at the initial phase of eggshell mineralization (5, 6, and 7 h p.o., when the nucleation sites appear and early mineralization starts), at the beginning of the active calcification phase (10 h p.o.) or during the linear growth phase of rapid calcification (16 h p.o.). Various tissues (D, K, and L) and oviduct regions (Ma, WI, RI, and Ut) were collected from birds at each stage of shell calcification. Mid-shaft tibial bone (B) from 90-week-old hens (18 h p.o., at the end of bone resorption) were also harvested. Tissues were directly immersed in liquid nitrogen and were kept at Ϫ80°C.  a to b) indicates the mapping zone across which EDS analysis was performed to obtain elemental spectra along the line for the indicated elements (B). C, spectrum 1 corresponds to EDS analysis of background in the absence of vesicles, and spectrum 2 and spectrum 3 correspond to EDS analysis at two different locations within a vesicle, as indicated in A. UFs were negatively stained with uranyl acetate 2%. Intensities are given in arbitrary units.

EDIL3/MFGE8 in avian shell calcification Soluble eggshell matrix and tissue protein extraction
Eggshell matrix proteins from eggs were extracted as detailed previously (107). Briefly, eggshell pieces with eggshell membranes were immersed in 154 mM NaCl solution containing protease inhibitors (2.5 mM benzamidine-HCl, 50 mM ⑀-aminon-caproic acid, 0.5 mM N-ethylmaleimide, and 1 mM phenylmethylsulfonyl fluoride) and ground into a fine powder. Eggshell powders were fully demineralized by immersion in 20% acetic acid with overnight stirring at 4°C. The resulting suspensions were dialyzed (cutoff at 3500 Da) against demineralized water during 24 h at 4°C and lyophilized. Samples were then incubated overnight at 4°C in 4 M guanidine HCl, 5 mM benzamidine HCl, 0.1 M ⑀-amino-n-caproic acid, 10 mM EDTA, 50 mM sodium acetate, and 1 mM phenylmethylsulfonyl fluoride. Samples were then dialyzed (cutoff at 3500 Da) against 0.5 M sodium acetate, pH 7.4, solution during 24 h at 4°C and then centrifuged for 10 min at 2000 ϫ g at 4°C. The resulting supernatants were stored at Ϫ20°C.

Electrophoresis and Western blot analysis
Protein concentrations were determined in various samples (SEM: Ut 10 h, UF 9 h, and D 10 h), using Bio-Rad DC Protein assay kit II (Bio-Rad, Marnes-la Coquette, France) in accordance with the manufacturer's instructions and using BSA as a standard. Fifteen micrograms of each sample (SEM: UF, Ut, and D) were diluted into Laemmli sample buffer (5:1), as described previously, and boiled for 5 min. Protein samples were then separated on 12% polyacrylamide gels (Mini-Protean II electrophoresis cell, Bio-Rad, Marnes-la-Coquette, France) and transferred onto 0.

Gene expression analysis
Total RNA was extracted from frozen samples (B, D, K, L, Ma, WI, RI, and Ut). RNAs from K, D, Ma, WI, RI, and Ut were extracted using NucleoSpin RNA commercial kit (Macherey-Nagel, Düren, Germany), and the B total RNA was extracted using RNA Now (Ozyme, Montigny-le-Bretonneux, France) and tandem TRI Reagent-TRI Reagent LS commercial kits (Sigma-Aldrich, Saint-Quentin Fallavier, France). Tissues were extracted at 10 h p.o., except for tibial bone, as this sample was solely collected at 18 h p.o. Additionally, Ut and RI samples were extracted at all stages (5-7, 10, and 16 h p.o.). Total RNAs from each sample were treated with TURBO DNA-free kit (Invitrogen). RNA concentration was measured at 260 nm, and their integrity was assessed on 1.5% agarose gels. Total RNA samples (1 g) were reverse-transcribed using RNase H-Moloney murine leukemia virus reverse transcriptase (Superscript II, Invitrogen) and Oligo(dT) TM primers (Invitrogen).

Synteny and phylogeny of EDIL3 and MFGE8
Synteny of edil3 and mfge8 genes were analyzed using the Genome data viewer of NCBI for eight vertebrate species (three mammals, two birds, one nonavian reptile, one amphibian, and two fishes).
EDIL3 and MFGE8 protein sequences of 12 vertebrate and 7 nonvertebrate species (Table S3) were aligned using Clustal Omega multiple alignment in MEGA7 (108). The appropriate substitution model, determined using MEGA7 substitution model, was the LG model. The phylogenetic tree was built with MEGA7 using Maximum Likelihood method (LG substitution model) with 1000 repetitions and the bootstrap method to evaluate branch robustness. Additionally, phylogenetic analysis using Bayesian inference was performed using BEAST version 1.10.4 (109) with 10,000,000 generations and a sampling of tree each 1000 steps. Trees were then summarized with TreeAnnotator (distributed with BEAST version 1.10.4) and displayed using FigTree version 1.4.4 (115) (http://tree.bio.ed.ac.uk/ software/figtree/). 3 Divergence times between species were determined using Timetree (www.timetree.org/) 3 (113) and are reported on the generated phylogenetic tree.

G. gallus EDIL3 and MFGE8 secondary and tertiary structure homology
Secondary structure of G. gallus EDIL3 (XP_424906.3) and MFGE8 (NP_001264039.1) peptide signal regions was predicted using PSIPRED version 3.3 (http://bioinf.cs.ucl.ac.uk/ psipred/). 3 EGF-like and F5/8C sequence regions were then loaded on the I-TASSER program server (111), to generate secondary and tertiary structural predictions. This template-based modeling program computed protein 3D structure predictions by molecular homology for each region of EDIL3 and MFGE8. The 3D structure prediction models were edited on PyMOL Molecular Graphics system.

Purification of extracellular vesicles from uterine fluid, examination by transmission EM, and energy-dispersive X-ray analysis
Immediately after dilution, EVs were isolated from UF harvested at 7 h p.o., using previously described methodology (112). Briefly, diluted UFs were centrifuged at 100 ϫ g for 15 min and then at 12,000 ϫ g for 15 min to remove cell debris (Sorvall ST40 R, Thermo Fisher Scientific, Waltham, MA). Two successive ultracentrifugation steps at 100,000 ϫ g (Beckman Coulter L8 -70M, Beckman, Fullerton) were then performed during 90 min to pellet the EVs. The pellet was suspended in 50 l of PBS and stored at Ϫ20°C until TEM analysis.
Ten microliters of EV fraction were deposited on an electron microscope grid (Copper ϩ Carbon) and then dried for 1 min. Excess liquid was absorbed with filter paper, and EVs were