Proteins involved in embryo-maternal interaction around the signalling of maternal recognition of pregnancy in the horse

During maternal recognition of pregnancy (MRP), a conceptus-derived signal leads to the persistence of the corpus luteum and the maintenance of gestation. In the horse, the nature of this signal remains to be elucidated. Several studies have focused on the changes in gene expression during MRP, but little information exists at the protein level. The aim of this study was to identify the proteins at the embryo-maternal interface around signalling of MRP in the horse (day 13) by means of mass spectrometry. A distinct influence of pregnancy was established, with 119 proteins differentially expressed in the uterine fluid of pregnant mares compared to cyclic mares and with upregulation of several inhibitors of the prostaglandin synthesis during pregnancy. By creating an overview of the proteins at the embryo-maternal interface in the horse, this study provides a solid foundation for further targeted studies of proteins potentially involved in embryo-maternal interactions, MRP and pregnancy loss in the horse.


Results
Sampling. Only reproductively sound mares with negative bacteriology and cytology of the uterine fluid were used for the sampling. In two cycles, namely one pregnant (P) and one control cyclic (C) cycle, a double ovulation occurred. Response to hCG resulted in ovulation 24-36 h after administration. In four cycles, ovulation only occurred 3 days after hCG; once in a P cycle, where artificial insemination (AI) was performed at the same time and in this case, the mare was inseminated a second time 48 h after the first time and she ovulated the day after. In all other P cycles, ovulation occurred within 48 h after AI. In one mare, a line of fluid was noticed by ultrasound of the uterus 1 day after AI and she was treated by intramuscular administration of oxytocin.
Identification of proteins. The average protein concentration was similar in the uterine fluid (UF) of P (9.2 g/mL) and C (9.8 g/mL) mares, while the average protein concentration in the yolk sac (YS) was only 78 µg/mL. For the first time, an overview was created of the proteins present in the UF and the YS at day 13 after ovulation in the horse. In the UF samples, a total of 10489 peptides were identified, accounting for 41% of all peptide like ions. Protein identification resulted in 1153 identifiable proteins (Supplementary file 1). After filtering and normalization, a total of 707 normalized proteins with at least two unique peptides were assessed for differential expression.
Differential expression of proteins was assessed for P versus C mares and pregnancy was associated with upregulation of 62 proteins ( Table 1) and downregulation of 57 proteins (Table 2). For all proteins in this comparison, the log fold change, the adjusted p-value and the number of peptides are listed in Supplementary file 3. In the YS samples, a total of 6500 peptide ions were identified, representing 51% of all peptide like ions and resulting in 903 identifiable proteins (Supplementary file 2). For the YS proteins, the primary goal was identification, rather than quantification, as different nature of the fluids impedes assessment of differential expression of proteins in YS versus UF.
Gene Ontology enrichment and pathway analysis. Categorization in the Gene Ontology (GO) terms 'molecular function' , 'biological process' and 'cellular component' is provided for all quantified proteins in the comparison of P versus C in Supplementary file 3. Figure 1 summarizes the GO categories in which the differentially expressed proteins are involved. The main category to which most proteins are assigned is 'cellular process (GO:0009987)' for the biological processes and 'binding (GO:0005488)' for the molecular functions. This coincides with the results in porcine uterine fluid 24 , but these are also the major categories when all proteins are taken into account. Overall, the differences in categorization between the groups are small.
Gene Ontology (GO) enrichment revealed no statistical overrepresentation when a Bonferroni correction for multiple testing was used (FDR < 0.05). No up-or downregulated KEGG pathways were detected either at a Benjamini-Hochberg corrected p-value of 0.05.

Embryo-maternal interaction.
Comparison of the proteins identified in the UF of the P mares and in the YS of the corresponding embryo revealed 347 common proteins, 806 proteins which were only detected in the UF and 556 proteins which were only found in the YS. Figure 2 represents an overview of these UF specific proteins, YS specific proteins and common proteins, with specific display of the proteins up-and downregulated during pregnancy and of the proteins categorized in the extracellular space.
A list of the 347 common proteins is provided in Supplementary file 4, including the functions in which these proteins are involved. Figure 3 summarizes the GOs in which these common proteins were involved. Similar to the results for the UF in Fig. 1, the main GO categories in which the common proteins are involved are also 'cellular process (GO:0009987)' and 'binding (GO:0005488)' and differences in categorization are small. Common proteins in YS and UF which were also upregulated in P versus C, showed a higher representation in the biological

Discussion
Maternal recognition of pregnancy is an intriguing subject in the horse and extensive research on the molecular processes involved has been performed in the field of transcriptomics 7,18 . However, information on the downstream translation to proteins is scarce. In this study, quantitative proteomics of the uterine luminal fluid assessing the effect of pregnancy was performed for the first time in the horse. At the same time, proteins in the embryonic yolk sac fluid were mapped to provide insight into the embryo-maternal interaction.
With 119 proteins differentially expressed in the uterine fluid of P versus C mares, a distinct influence of pregnancy was established. In general, a function of more than 40% of the differentially expressed proteins in the UF was categorized as 'binding (GO:0005488)' , coinciding with the findings in pigs and cattle, where the majority of proteins were also allocated to molecular binding 24,28 (Fig. 1). 'Binding' also represents the main category to which the common proteins in UF and YS were allocated, with subtly higher representation of proteins upregulated during pregnancy in categories linked to embryo-maternal interaction, namely 'developmental process (GO:0032502)' , 'response to stimulus (GO:0050896)' , 'transporter activity (GO:0005215)' and 'transcription factor activity -protein binding (GO:0000988)' at the expense of the more general GO term 'structural molecule activity (GO:0005198)' (Fig. 3). Cellular component categorization allocated 45% of the identified UF proteins to the extracellular space (Fig. 2). This coincides with the findings of Swegen et al. 34 , who specifically targeted secreted proteins by analysing embryo-conditioned medium. This supports the fact the proteins detected in our study mainly represent the proteins secreted in the uterine fluid rather than endometrial cells shed in the uterine lumen. This also accounts for the proteins which were found to be differentially expressed during pregnancy. Sixty four % of these proteins were categorized in the extracellular space; the other may have originated from occasional shedding of embryonic and endometrial cells into the uterine lumen. Figure 2 represents an overview of all UF specific, YS specific and common proteins, including their differential expression in P vs C and their allocation to the extracellular space. Interestingly, the majority of proteins commonly found in UF and YS are indeed present in the extracellular space. These represent candidate proteins involved in embryo-maternal interaction and signalling. In general, our results greatly coincided with the findings of Swegen et al. 34 who worked with day 8 blastocysts to examine proteins present in and secreted by early equine embryos. Figure 6 shows the number of proteins which were commonly found in the blastocoel fluid and the YS and those found to be secreted in embryo-conditioned medium at 24 h and 48 h and in the UF in our study. More than two third of the proteins reported in the blastocoel fluid were also detected in the YS and more than one third of the proteins found to be secreted after 48 h of embryo culture were also detected in the UF. Overlap of the results validates our independent findings on the one hand and indicates conserved expression of several proteins throughout development on the other hand.
In the context of MRP, prostaglandin synthesis is of special interest. For three proteins involved in this pathway, namely prostaglandin reductase 1 (PTGR1), glutathione transferase 1 (GSTP1) and annexin A1 (ANXA1), significantly higher amounts were detected in the uterine fluid of pregnant mares compared with cyclic mares. Apart from acting on 15-oxo-PGE1, 15-oxo-PGE2 and 15-oxo-PGE2-alpha as 15-oxo-prostaglandin 13-reductase, PTGR1 catalyzes leukotriene B4 into its biologically less active metabolite, being the key step in the metabolic inactivation of leukotriene B4, as depicted in Fig. 7.

Protein Symbol
Log FC Adj. p-value Gene Symbol Gene Description  While glutathione transferases are also generally involved in the biosynthesis of prostaglandins and leukotrienes, as well as progesterone and testosterone 36 , a specific anti-inflammatory effect of GSTP1 by reduction of PTGS2, formerly known as cyclooxygenase-2 (COX-2), has been described 37 . Furthermore, transport of GSTP1 across the plasma membrane was demonstrated 37 . Based upon these observations with recombinant human GSTP1 in mice and the high homology of equine GSTP1 with other species, equine GSTP1 in uterine fluid might cross the plasma membrane and target intracellular PTGS2. Interestingly, GSTP1 was also detected in the YS of the equine conceptuses. In this regard, the pregnancy associated upregulation of GSTP1 observed in the equine uterine fluid could be involved in the luteostatic mechanism by inhibiting PTGS2. Further research is needed to examine this hypothesis, as this is the first report on the presence of GSTP1 in equine uterine fluid.
Another anti-inflammatory factor with an inhibitory effect on prostaglandin synthesis, more specifically on phospholipase A2, is annexin A1 (ANXA1) 38,39 . Annexin A1 was upregulated in the uterine fluid of the pregnant mares when compared to the cyclic mares and this association of annexins with pregnancy coincides with literature. An increase in ANXA1 was also reported in the uterine luminal fluid of pregnant ewes from day 10 to day 12 25 . Several annexins have been linked to embryo-maternal interaction. Annexin 4 (ANXA4) was found to increase over time from day 10 to day 13 in both cyclic and pregnant pigs 24 , and we previously reported greater quantities of ANXA4 in the oviductal fluid of pregnant mares, when compared to cyclic mares 40 . In our study, we detected annexin 1, 2, 3, 4, 5, 7 8 and 11 in the UF, while ANXA2 and ANXA5 were also found in the YS. Swegen et al. 34 also reported the presence of ANXA2 in both the equine blastocoel fluid and the embryo-conditioned medium after 48 h. The only annexin found to be upregulated during pregnancy was ANXA1. Annexin 1 is an inhibitor of phospholipase A2, a rate-limiting enzyme which liberates arachidonic acid for the synthesis of prostaglandins and for which a lower enzyme activity of phospholipase A2 has been demonstrated in pregnant mares compared to cyclic mares on day 14 41 . In our study, phospholipase A2 group IIA (PLA2G2A) tended to be downregulated in the uterine fluid of pregnant mares with a logFC of −1.31 compared to the cyclic condition, but it was not significant at a 0.05 FDR. Both PLA2G2A and phospholipase A2 group VII (PLA2G7) were detected in the YS; the latter was also found in the equine embryo-conditioned medium after 48 h 34 . Overall, our data suggest pregnancy associated interference with the luteolytic eicosanoid pathway with upregulation of inhibitory factors at different levels of the prostaglandin synthesis pathway.
A close interaction between prostaglandins and oxytocin has been described in the context of MRP in the horse with downregulation of the oxytocin receptor protein in the pregnant endometrium on day 14 13 . In the present study, the presence of oxytocin in the uterine luminal fluid was examined, but it was not detected. However, this does not mean it was not present in the original samples; the collection method might have retained some peptides and missing values are intrinsic to mass spectrometry 42,43 . Phosphoinositide phospholipase C (PLCD1), involved in the oxytocin receptor signalling pathway, was detected, but not significantly affected by pregnancy 44 . The reduced expression of OXTR during pregnancy has been hypothesized to be induced by an observed decrease in the gene expression of oestrogen receptor 1 (ESR1) in the pregnant equine endometrium 15 . Several proteins related to ESR1 were also found to be affected in the uterine fluid. Surprisingly, pregnant mares showed a strong upregulation of deoxynucleotidyltransferase terminal interacting protein 2 (DNTTIP2), previously known as oestrogen receptor binding protein (ERBP). Binding of DNTTIP2 to ESR1 enhances its transcription 45 . Upregulation of DNTTIP2, which would lead to increased transcription of ESR1 in pregnant mares is contradictory to findings in literature and further targeted research is required to clarify this aspect. Downstream of the ESR1, the influence of oestrogen on the ezrin-radixin-moesin (ERM) family of actin-binding proteins has been studied, mainly in the context of breast cancer 46,47 . The distribution pattern of ERM-proteins in the blastocyst and the uterus has been linked to the implantation potential in mice. Protein analysis of uterine fluid has demonstrated the presence of ezrin (EZR) and moesin (MSN) in cattle 26,27 and pigs 23 . In the horse, upregulation of EZR and downregulation of MSN was detected, while an inverse association with pregnancy was noted in cattle 26,27 . We detected both EZR and MSN in the YS and they were also found in the blastocoel fluid 34 .
Apart from the specific interest in proteins involved in prostaglandin synthesis, we also aimed to create a general overview of the proteins present at the embryo-maternal interface and potentially involved in signalling. Supplementary File 4 presents all proteins which were commonly found in the UF of P mares and in the YS and the functions of each protein are included. To visualize their role in embryo-maternal interaction and signalling, the proteins involved in GO terms including 'embryo' , 'maternal' or 'uterus' are depicted in Fig. 4 and those linked to GO terms 'growth factor' and 'cytokine' in Fig. 5. The origin of the proteins can be distinguished in red (UF), yellow (YS) and pink (UF and YS) and up-and downregulation during pregnancy is represented by enlargement or shrinkage of the protein respectively. Interestingly, most proteins which were found to be upregulated during pregnancy were detected both in UF and in YS, while downregulation during pregnancy generally coincided with absence of these proteins in the YS, indicating a potentially important role of the embryo in the production of these proteins during pregnancy. Several common proteins were found at the embryo-maternal interface during MRP in cattle, including aconitase 1 (ACO1), which was specifically detected in the uterine fluid of pregnant and not in cyclic heifers, as well as glucose-6-phospate isomerase (GPI), which has been detected in the uterine fluid of both pregnant and cyclic heifers and for which an embryonic source has been presumed based on transcriptomics 27 . In our study, both proteins were found in the YS and the UF of P mares, with significant upregulation of GPI in P versus C. Two other proteins which were commonly found in UF and YS, namely FK506 binding protein 4 (FKBP4) (Fig. 4) and heat shock protein 90 (HSP90AB1) (Fig. 5), have been elaborately discussed by Swegen et al. 34 concerning their progesterone supportive role. Co-operation of both factors is necessary for activation of the progesterone receptor 48 , FKBP4 has shown to be crucial for uterine receptivity and implantation in mice 49 and FKBP4 deficit has been associated with pregnancy loss in human 50 . While FKBP4 was detected in equine blastocoel fluid and speculated to be involved in signalling, it was not detected in the embryo-conditioned medium 34 . Interestingly, we did find both FKBP4 and HSP90AB1, not only in YS, but also in UF, even though their presence was not affected by pregnancy. While further confirmation of the role of specific proteins is required, the overview created in this study can be used as a basis for further targeted studies in the horse.
In addition to the role in prostaglandin and progesterone metabolism, involvement in proteolysis and lipid metabolism was also prominent in the commonly detected proteins in our study and the one of Swegen et al. 34 , also coinciding with previous findings on transcriptomics around MRP 15 . Several cathepsins (G, D, L and S) were detected in the UF with downregulation of cathepsin L (CTSL) and S during pregnancy. Pregnancy associated downregulation of CTSL1 was also found at the transcriptome level 15 . Considering lipid metabolism, we detected differential expression of lipocalin (P19), apolipoprotein A1 (APOA1) and apolipoprotein D (APOD). These proteins are important transporters of essential lipids to the developing conceptus. Retinol binding protein (RBP), which also belongs to the lipocalin family, and APOA1 have been detected in uterine fluid of pregnant and cyclic pigs, cattle and sheep [23][24][25]27,28 , with increasing amounts between day 10 and day 13 in both pregnant and cyclic pigs 24 . In the horse, lipocalin (P19), apolipoprotein A1 (APOA1) and apolipoprotein D (APOD) were all downregulated in the uterine fluid of the pregnant mares. Pregnancy associated upregulation of APOA1 was reported at the transcriptome level 15 and presence of P19 and APOA1 in the yolk sac fluid illustrates their role in the embryo-maternal dialogue. Therefore, lower amounts in the uterine fluid during pregnancy rather indicate the transport and binding to the conceptus. Lipocalin P19 or uterocalin is a progesterone induced protein, which is abundantly present in the equine uterine secretions during dioestrus and early pregnancy 51,52 . While the early developing equine conceptus moves around the uterus, it entirely depends upon the uterine secretions for its nutrition and P19 can function as a carrier for essential lipids and amino acids 53 . Coinciding with our findings, P19 has been detected in the trophoblast and the yolk sac fluid of the equine embryo 51,52 and it is one of the most abundant proteins in the embryonic capsule [54][55][56] . Therefore, the lower amount of P19 in P versus C is probably due to binding of substantial quantities to the embryo.
While a novel and informative overview is created, it has to be borne in mind that no statistically significant results were obtained at the level of molecular functions, biological processes and pathways. Differential expression of individual proteins was observed between the different UF conditions, and these proteins were categorized in GO terms, but statistical analysis showed no significant overrepresentation of any of the GO terms or KEGG pathways. Furthermore, it should be noted that MS intrinsically suffers from missing values and conclusions based on the absence of proteins cannot be made 42,43 . However, the field of proteomics has greatly evolved in recent years, providing the possibility for statistically robust quantitative comparison of individual protein levels in complex biological samples, like uterine fluid 22 . HDMSE specifically has been shown to provide good proteome coverage and reproducibility 35 . At the same time, however, analysis of GO terms and pathways for proteomics is still in its infancy 57,58 . As many of the here described bioinformatics approaches for proteomic analysis were originally developed for genomics, a similar but more matured field, their performance can be expected to show a similar growth as that of the genomic approaches. Moreover, the similarity between these fields potentially allows an integrated approach in which results from several omics studies can be combined. In conclusion, proteins present in the equine uterine and embryonic yolk sac fluid around the signalling of MRP at day 13 were identified and quantified at large scale for the first time in the horse. We detected upregulation of several inhibitors of prostaglandin synthesis, including PTGR1, GSTP1 and ANXA1, in the uterine fluid of pregnant mares. Overall, an overview was created of the proteins playing a role at the embryo-maternal interface in the horse. This study provides a solid foundation for further targeted studies of proteins potentially involved in embryo-maternal interactions, maternal recognition of pregnancy and pregnancy loss in the horse.

Methods
Sampling. All animal handlings were approved by the Ethical Committee of the Faculty of Veterinary Medicine (EC2013/118) of Ghent University. All methods were performed in accordance with the relevant guidelines and regulations. A switch back design was followed with 5 mares undergoing two different types of cycles: a pregnant cycle (P) and a cyclic control cycle (C). In this way, the samples were paired using the same mare as its own control for pregnancy and the experimental unit was the mare. The order of P and C cycles was randomly altered for the different mares. No resting cycles were included. During the breeding season, five reproductively sound Warmblood mares between 4 and 13 years old were monitored by transrectal ultrasound. Reproductive soundness was confirmed by negative cytology and bacteriology. Mares displaying uterine oedema and a follicle exceeding 35 mm received 1500 IU hCG intravenously and were either inseminated the next day with fresh semen of the same stallion (P) or left unbred (C). Ovulation was evaluated twice daily by ultrasound. In both groups, sampling was performed 13 days after detection of ovulation. To recover undiluted uterine fluid in order to avoid negative effects of excessive Ringer's salts on MS 59 , intra-uterine application of a tampon (OB Mini; Johnson & Johnson, Beerse, Belgium) was performed based upon the method described by Wolf et al. 33 . A double gloved technique was used to avoid vaginal contamination. The tampon was left in the uterus during 10 minutes and upon removal it was placed in a Falcon tube at 4 °C until further processing. Subsequently, the mare's uterus was flushed with sterile Ringer's solution by means of a modified endotracheal tube to recover the embryo (P).
To process the uterine fluid, 1 mL of sterile water (B60, Biosolve, Valkenswaard, The Netherlands) was infused on top of the tampon and the tampon was attached in the upper part of the Falcon tube by fixing the cord with the cap. Subsequently, the Falcon tube was centrifuged for 20 minutes at 1000 × g at 4 °C. The supernatant was collected and stored in a Protein LoBind Eppendorf tube (Eppendorf AG, Hamburg, Germany) at −80 °C. Meanwhile, the embryo was isolated in a petri dish and the yolk sac fluid was collected by aspiration with a 21 G needle and stored in a Protein LoBind Eppendorf at −80 °C.
A total of 15 samples were collected, consisting of uterine fluid (UF) (n = 10) from five biological replicates coinciding with the five mares (1-5) for the P and C treatment cycles, as well as yolk sac fluid (YS) (n = 5) from the P cycles.
Sample preparation for mass spectrometry analysis. After thawing, protein concentration in each sample was determined using the Coomassie (Bradford) Protein Assay Kit (Thermo Fisher Scientific, San José, CA, USA) according to the manufacturer's instructions. Further processing was performed for 10 µg protein of each uterine fluid sample and for 500 ng protein of the yolk sac samples. Samples were dissolved in 20 μL 0.5 M triethylammonium bicarbonate (TEABC; Sigma-Aldrich, St. Louis, MO, USA). Two µl of reducing agent (10 µM DTT; Invitrogen, Merelbeke, Belgium) were added followed by incubation for 1 h at 60 °C. Subsequently, 1 µl of alkylizing agent (200 mM methyl methanethiosulfonate (MMTS) in isopropanol; Sigma-Aldrich, St. Louis, MO, USA) was added and samples were incubated for 10 min at room temperature. Digestion was performed overnight at 37 °C with trypsin lys C (1:20, trypsin:protein w/w, Promega, Leiden, The Netherlands) in TEABC buffer with 1 mM CACL(2) and 5% acetonitrile (Biosolve, Valkenswaard, The Netherlands). Samples were vacuum-dried and stored at −20 °C until analysis.
Data acquisition by HDMS E analysis. The peptides were separated using a nanoscale UPLC system (nanoAcquityUPLC, Waters, Milford, USA) coupled to a Synapt G2-Si mass spectrometer (Waters). Peptides were first trapped in 0.1% formic acid on a 180 µm × 20 mm C18 Trap column. Separation was performed on a HSS C18 1.8 m, 100 m × 250 mm analytical column at a flow rate of 300 nL/min and a temperature of 45 °C. As mobile phase A a 0.1% formic acid with 4% DMSO in water solution was used and 80% ACN containing 0.1% formic acid constituted mobile phase B. Peptides were separated for 60 min at 1-40% solvent B and for 1 min Figure 5. Involvement of proteins found in the yolk sac fluid and the uterine fluid of pregnant mares in GO terms and pathways representing embryo-maternal interaction. All GO terms and pathways that include 'cytokine' or 'growth factor' in their description were selected, together with all identified proteins in either the yolk sac or uterine fluid belonging to these GO terms or pathways. These GO terms, pathways and proteins were then visualized using Cytoscape 3.3.0. Proteins found only in the uterine fluid are represented as red circles, proteins found only in the yolk sac as yellow circles and proteins found in both as purple circles. Proteins significantly (FDR corrected p-value < 0.05) up-or downregulated in the uterine fluid of pregnant mares are respectively larger and smaller circles (size not scaled with magnitude of up-or downregulation). GO terms and pathways are represented as a blue 'V' , with lines indicating whether a GO term or pathway is associated with a protein.
40-85% solvent B. Seven minutes of rinsing (85% solvent B) re-equilibrated the column to the initial conditions. Eluted peptides were analysed in positive mode ESI-MS using High Definition MS E (HDMS E ) with a collision energy look up table as described in 22 . The spectral acquisition time of low and elevated energy scans was 0.6 s over an m/z range of 50-2000. [Glu1]-Fibrinopeptide B was used for post-acquisition lock mass correction. All  34 . Figure 6A shows the number of proteins which were found in the blastocoel fluid by Swegen et al. 34 and the YS in our study and Fig. 6B illustrates the proteins those to be secreted in embryoconditioned medium at 24 h and 48 h by Swegen et al. 34 and in the UF in our study. Numbers are based on the reported gene symbols. UF samples were analysed in the same run; three technical replicates (R1-R3) were run for each sample and four quality controls (QC) were included in which all samples were pooled.
Identification and quantification of peptides and proteins. All data were processed in Progenesis QIP (Progenesis QIP 2.0, Nonlinear Dynamics, Waters), including normalization and quality control. A database with UniProt IDs was created by conversion of Ensembl gene identifiers for Equus caballus (n = 22295) to Uniprot IDs using http://www.uniprot.org/uploadlists/ and including common contaminants (http://www. thegpm.org/crap/). As only secreted proteins are expected to be found, it can be argued that this database should be limited to only these secreted proteins. However, there is much debate on the accuracy of FDR calculations with such limited databases [60][61][62] and as such a cautious approach was taken in which all proteins were assessed. Using Progenesis QIP, peptides were identified against this database with a FDR of 4% 63 and allowing maximum one miscleavage. Protein quantification was based on the Hi-3 method 64 , which uses the average of the three most intense peptides of each protein for its quantification. Resulting normalized abundances for each protein, as well as unique peptide counts were further used for analysis of differential expression.
Analysis of differential expression. Analysis of differential expression was performed for the UF samples. Only normalized abundancies of proteins with at least two unique peptides (n = 707) were included in the analysis. Pairwise comparisons of differential expression were made for P versus C with the individual horses as a blocking factor, using R Bioconductor limma package 65 and a FDR of 0.05.
Gene Ontology enrichment and pathway analysis. Gene Ontology (GO) terms (molecular functions, biological processes and cellular locations) were downloaded for each protein with the PANTHER Classification System 66 . For the pair-wise comparison of P and C, a statistical overrepresentation test against all quantified proteins was performed for all significantly up-and downregulated (FDR < 0.05) proteins. These tests were done for all primary GO classes: molecular functions, biological processes and cellular components. Pathways were analysed with Bioconductor's 67 GAGE package 68 . LogFC values of all quantified proteins were used as input against Equus caballus background reference pathways from KEGG.
Proteins involved in the embryo-maternal interaction were visualized using Cytoscape 3.3.0. To visualize embryo-maternal signalling, all GO terms and pathways that include 'cytokine' or 'growth factor' in their description were selected, together with all identified proteins in either the yolk sac or uterine fluid belonging to these GO terms or pathways. The connection between these GO terms, pathways and proteins was then visualized using Cytoscape 3.3.0. The same methodology was used to create a network based on GO terms and pathways including 'embryo' , 'maternal' or 'uterus' in their description. Data availability. All data are available in the Supplementary files.