Aggregation of Omic Data and Secretome Prediction Enable the Discovery of Candidate Plasma Biomarkers for Beef Tenderness

Beef quality is a complex phenotype that can be evaluated only after animal slaughtering. Previous research has investigated the potential of genetic markers or muscle-derived proteins to assess beef tenderness. Thus, the use of low-invasive biomarkers in living animals is an issue for the beef sector. We hypothesized that publicly available data may help us discovering candidate plasma biomarkers. Thanks to a review of the literature, we built a corpus of articles on beef tenderness. Following data collection, aggregation, and computational reconstruction of the muscle secretome, the putative plasma proteins were searched by comparison with a bovine plasma proteome atlas and submitted to mining of biological information. Of the 44 publications included in the study, 469 unique gene names were extracted for aggregation. Seventy-one proteins putatively released in the plasma were revealed. Among them 13 proteins were predicted to be secreted in plasma, 44 proteins as hypothetically secreted in plasma, and 14 additional candidate proteins were detected thanks to network analysis. Among these 71 proteins, 24 were included in tenderness quantitative trait loci. The in-silico workflow enabled the discovery of candidate plasma biomarkers for beef tenderness from reconstruction of the secretome, to be examined in the cattle plasma proteome.


Introduction
Animal products are the main source of protein and essential nutrients in human nutrition. While in developing countries, the objective is to increase meat production to meet human nutritional needs, in industrialised countries the major expectations concern meat quality [1] A challenge for the beef sector in those countries is to predict and manage the meat quality attributes in order to ensure their low variability. Among the attributes of beef eating quality (tenderness, juiciness, flavour and colour), tenderness is a top priority for the beef industry to meet consumers' expectations [2] However, beef tenderness is a complex phenotype with large individual variation within and between animals that can vary according to multi-factorial influences. Factors related to the animal itself including genotype [3] and physiological type (breed, age, and sex) [4][5][6] contribute to the variability in tenderness. Extrinsic factors include management systems and rearing conditions [7][8][9], animal transport and handling during the pre-slaughtering period, slaughtering conditions [10], and post-slaughter factors including maturation, storage and cooking [4,11].
Today, meat tenderness attributes are assessed only after animal slaughtering and meat ageing which limits the delivery of consistent quality meat [12][13][14]. Thus, the identification of biomarkers for meat quality measurable in living animals is a good opportunity to develop monitoring, decision-making and management tools for beef quality prior to slaughter. Thanks to genomics, several research groups have investigated the potential of muscle-derived markers for characterizing the molecular mechanisms underlying beef tenderness as well as for prediction purpose. Some DNA polymorphisms and transcript abundances were related to variation in tenderness. Thus, markers linked to genetic polymorphism were identified in proteolytic genes e.g., CAPN1, CAST [15,16] and marketed as genetic tests. Transcriptional muscle profiling enabled the detection of gene transcripts involved in fat, energy metabolism and heat shock response (e.g., DNAJA1, HSPB1 and CRYAB), as candidate biomarkers for meat tenderness [17,18], which were included in a dedicated micro-array [18]. The development of proteomics has taken the issue of identification of tenderness biomarkers a step further [19,20]. Proteomic studies confirmed the importance for meat tenderness of proteins involved in muscle structure, energy metabolism, proteolysis or apoptosis (for a review, Picard et al. [21]). However, a high variability in muscle biomarker content is detected among breeds, individuals and muscles [22]. In addition, inverse relationships between some biomarkers and beef tenderness were also reported as a function of muscle properties [23].
So far, biomarker assessment requires muscle sampling in slaughtered animals or biopsies on living animals. Thus, the identification of generic and low invasive biomarkers in body fluids is an issue for molecular phenotyping in living animals [24]. As circulating proteins mirror the individual's physiology, identification of plasma biomarkers could allow prediction of the tenderness potential of living animals. In this study, we hypothesized that the aggregation of public data may help to identify candidate plasma biomarkers for beef tenderness from the secretome of muscle. We thus designed a workflow to generate a dataset of known biomarkers for tenderness and predict in silico the proteins secreted through conventional pathways or other pathways allowing transit of proteins from muscle to the plasma.

Literature Search and Data Aggregation
A total of 459 articles including one GSE were identified using the MEDLINE, GOOGLE and CLAVIRATE analytics as related to meat tenderness ( Figure 1). Among them, 425 articles were excluded because they did not meet the criteria of inclusion. From the corpus of the 44 remaining publications, 26 articles were identified as eligible for proteomic data [17,. Eleven articles including the series accession number GSE9256 (PMID: 18443416) were found as eligible for transcriptomic data [15,17,18,24,41,[48][49][50][51][52][53]. Twelve articles were found as eligible for genetic data [18,50,[54][55][56][57][58][59][60][61][62][63]. The computational data aggregation from these 44 publications gave an overview of 1299 ID gene name (GN) related to meat tenderness whatever the muscle, breed, animal type, sex, age at slaughter, geographic area, and methodologies used for tenderness evaluation. Depending on the type of molecule studied (protein, transcript or gene): 139 unique GN were reported as proteomic data, 249 unique GN as transcriptomic data, and 123 unique GN as genetic data. The compilation of these three lists generated the aggregated dataset comprising 469 non-redundant GN (Table 1, Figure 2).

Figure 1.
Flowchart of the workflow applied for the discovery of candidate plasma biomarkers for beef tenderness using a review of the literature and aggregation of omic data Figure 2. Origin of the omics data included in the study. The Venn diagram shows the intersects of the three omic datasets aggregated in the study. The aggregated dataset related to tenderness [64] was limited to the unique ID Gene Names.   Origin of the omics data included in the study. The Venn diagram shows the intersects of the three omic datasets aggregated in the study. The aggregated dataset related to tenderness [64] was limited to the unique ID Gene Names. Origin of the omics data included in the study. The Venn diagram shows the intersects of the three omic datasets aggregated in the study. The aggregated dataset related to tenderness [64] was limited to the unique ID Gene Names.  (44) The table presents the number of ID Gene Names for each dataset. Predictive secreted proteins (conventional pathways, (i): number of ID Gene Names identified as predicted secreted with signal-P sequence using ProteINSIDE predictive analysis (Signal-P > 0.5; Target-P ≤ 2). Predictive secreted proteins (unconventional pathways (UPS), (ii): number of ID Gene Names identified as predicted secreted without signal-P sequence using ProteINSIDE predictive analysis (Target-P ≤ 3). Remaining proteins: number of ID Gene Names non-predicted as secreted using ProteINSIDE. Predicted secreted proteins (conventional and alternative) in plasma: Number of ID Gene Names: (i) and (ii) found in the plasma by overlapping with the Bovine Proteome Atlas (BPA). Other proteins hypothetically secreted in plasma: Remaining proteins that were found in plasma by overlapping with the Bovine Proteome Atlas (BPA). In brackets: number of unique ID Gene Names associated with each category of proteins in the aggregated dataset. Unconventional pathways of secretion (UPS).

Computational Prediction
Prediction of the secreted proteins. Table 1 illustrates the numbers and characteristics of the proteins associated with the omics datasets. The predictive analysis using ProteINSIDE from the aggregated dataset allowed us to identify 54 proteins (11.5%) as predicted secreted proteins according to a conventional pathway (with signal-P and/or TM domain) and 36 proteins (7.7%) as predicted secreted proteins according to UPS pathways (without signal-P). The list of remaining proteins included 379 GN (80.8%).
Prediction of the secreted proteins putatively found in the plasma. The intersection of the datasets and the Bovine Plasma proteome Atlas (BPA) allowed to retrieve proteins that may be secreted by conventional or by UPS pathways and found in the plasma, and the remaining proteins not hallmarked for secretion but found in the plasma respectively (Table 1). Thirteen proteins referred to as "predicted secreted proteins in plasma" (2.8%) and 44 proteins referred to as "hypothetically secreted proteins in plasma" (9.4%) were identified respectively (Table 1). These repertoires are presented in Table 2.  We report all the proteins proposed as plasma candidates for beef tenderness: 13 predicted secreted proteins identified using ProteINSIDE tool, 44 hypothetically secreted found by overlapping the repertoire of proteins not hallmarked for secretion with the BPA, and 14 plasma proteins revealed from the network and QTL analysis. EVs: The vesicular proteins were retrieved by overlapping with the Vesicular protein Atlas from HPA. The exosome proteins were retrieved by overlapping with the Exosome proteins from Exocarta Atlas. BPA: Bovine Plasma proteome Atlas. The information on the location of the genes encoding proteins of interest within published QTL for tenderness retrieved using the ProteoQTL module of ProteINSIDE. This module interrogates a publicly available QTL library in Animal QTL database that contains cattle QTL and the published data associated. In brackets in the QTL column: chromosome associated with the Tenderness score and/or Shear force QTL. "X" means that the protein was found in the considered HPA and/or Exocarta atlas.

Gene Ontology
The full compiled atlas of 469 GN and the repertoires of 13 "predicted secreted proteins in plasma" and of 44 "hypothetically secreted proteins in plasma" were then submitted to Gene ontology (GO) annotation. The biological processes (BP) associated with the different datasets are presented in the Tables 3-5 respectively. The hierarchical "varonoi" visualization of the canonical pathways related to the 13 "predicted secreted proteins in plasma" and the 44 "hypothetically secreted proteins in plasma" are shown in Supplementary Data 1 and 2. A SimRel semantic rapprochement performed on the TOP50 of the GO terms associated with the 469 proteins (p-value adjusted <0.001, minimum of two proteins annotated in annotation) highlighted 10 BP: "Inflammatory response", "Gluconeogenesis", " Protein stabilization", "chaperone-mediated protein complex assembly", "Carbohydrate metabolism", "Aging", "Muscle contraction and development", "cell adhesion", "protein folding" and "Apoptotic process" (Table 3). Thanks to REVIGO semantic rapprochement performed on the GO terms associated with the 13 "predicted secreted proteins in "plasma" (p-value adj. <0.05, minimum of two proteins annotated in GO annotation), s BP were identified: "Cell adhesion", "Apoptotic process", "Endocytosis", "Response to oxidative stress", "Hydrogen peroxide metabolism" and "Lipid metabolism" (Table 4). In parallel, thanks to the Reactome visualization of the 13 "predicted secreted proteins"; four major canonical pathways were identified: "homeostasis", "signal transduction (receptor tyrosine kinase signaling, and NR1H2/H3 mediated signaling)", "immune system (neutrophil degranulation)" and "transport of small molecules (plasma lipoprotein assembly, remodeling, ABC transporter ion channel, mitochondrial calcium ion transport) (Supplementary Data 1). Thanks to semantic rapprochement performed on the GO terms associated with the 44 "hypothetically secreted proteins in plasma" (p-value adj. <0.001, minimum of two proteins annotated in GO annotation), 9 BP were identified: "Protein stabilization", "Gluconeogenesis", "response to ethanol", "Protein folding and chaperone-mediated protein complex assembly", "Endocytosis", "Muscle contraction", "Viral process" and "Hydrogen peroxide metabolism" (Table 5). In parallel, thanks to the Reactome visualization of the 44 "hypothetically secreted proteins in plasma"; 10 major canonical pathways were identified: "cell-cell communication", "homeostasis", "muscle contraction", "metabolism of proteins", "metabolism of lipids (citric acid cycle and carbohydrate metabolism)", "programmed cell death", "cellular responses to external stimuli", "organelle biogenesis and maintenance (cilium assembly . . . )", "autophagy", "extracellular matrix organization" (Supplementary Data 2). The comparison between the repertoires of 13 "predicted secreted proteins in plasma" and of 44 "hypothetically secreted proteins in plasma" revealed six common GO Biological Process including "receptor-mediated endocytosis", "cellular response to oxidative stress", "hydrogen peroxide catabolic process", "neutrophil degranulation", "oxidation-reduction process" and "cellular oxidant detoxification" (Figure 3).      We report the Top5O of the "Biological process" Gene Ontology terms identified with a significant p-value (p-value < 0.001) and associated with a minimum of two proteins. This GO Table was    We report all of the "Biological Process" terms associated with the Gene Ontology annotations identified with significant p-values (p-value < 0.05) and associated with minimum of two proteins. This GO Table was      We report all of the Biological Process associated with the Gene Ontology annotations identified with a significant p-values (p-value < 0.001) and associated with minimum of two proteins. This GO Table was

Network Analysis and Plasma PPi Identification
Examination of the network built from all of the 57 plasma candidates identified in this study (13 "predicted secreted proteins in plasma" and 44 "hypothetically secreted proteins in plasma" combined) revealed 544 interactors of which 75 proteins were present in the BPA (Figure 4). Eleven proteins out of the 57 plasma candidates (ATP5B, BPGM, COL11A1, COL13A1, ENO3, FGF12, LRRC16A, PCDH7, PGAM2, PVALB and TG) were not included in the MINT database used to generate the network from Cytoscape. Finally, the investigation of these 75 candidate proteins allowed to identify 14 additional proteins (CASP8AP2, ZBTB21, USP8, NEFL, CAT, GSS, PRKACB, CFL1, MAPK1, CCNB2, ACTN1, YWHAZ, YWHAB and PSMA7) that could be new meat tenderness proteins located in cattle meat Quantitative trait loci (QTL) for Shear force and/or Tenderness score ( Table 2). These 14 proteins were included in the repertoire of the "secreted proteins in plasma". Thus, a repertoire of 71 non-redundant candidate plasma proteins related to tenderness was generated ( Table 2).

Network Analysis and Plasma PPi Identification
Examination of the network built from all of the 57 plasma candidates identified in this study (13 "predicted secreted proteins in plasma" and 44 "hypothetically secreted proteins in plasma" combined) revealed 544 interactors of which 75 proteins were present in the BPA (Figure 4). Eleven proteins out of the 57 plasma candidates (ATP5B, BPGM, COL11A1, COL13A1, ENO3, FGF12, LRRC16A, PCDH7, PGAM2, PVALB and TG) were not included in the MINT database used to generate the network from Cytoscape. Finally, the investigation of these 75 candidate proteins allowed to identify 14 additional proteins (CASP8AP2, ZBTB21, USP8, NEFL, CAT, GSS, PRKACB, CFL1, MAPK1, CCNB2, ACTN1, YWHAZ, YWHAB and PSMA7) that could be new meat tenderness proteins located in cattle meat Quantitative trait loci (QTL) for Shear force and/or Tenderness score ( Table 2). These 14 proteins were included in the repertoire of the "secreted proteins in plasma". Thus, a repertoire of 71 non-redundant candidate plasma proteins related to tenderness was generated ( Table 2).  Figure 4. Network of the 71 plasma proteins identified in this study as putative candidate biomarkers for beef tenderness. This network reports the 71 plasma proteins identified as candidate biomarkers for meat tenderness in this study. The 13 predicted secreted proteins in plasma (conventional and alternative pathways) are shown in purple ellipse. The 44 hypothetically secreted proteins in plasma are shown in pink ellipse. The green rectangle refer to interactor identified through the up to date Cytoscape tool (MINT resource, Psciquic web service, 2017-12-17). The border red rectangle refer to the interactors located in cattle meat QTL for Shear force and/or Tenderness score tenderness (ProteINSIDE ProteoQTL analysis). Solid line shows the "primary interaction type". Dotted line shows the interaction through "detection method". Eleven out of the 57 plasma candidates (ATP5B, BPGM, COL11A1, COL13A1, ENO3, FGF12, LRRC16A, PCDH7, PGAM2, PVALB and TG), not included in the MINT database, are not shown in this network.

Identification of the Extracellular Vesicles (EVs) Proteins
The overlapping of the 71 plasma candidates with the vesicular proteins atlas (HPA) and the Exosome protein atlas (Exocarta) respectively allowed identifying several proteins likely to be secreted through EVs pathways. Thus, 13 vesicular proteins (ACTB, ALB, APOE, FASN, FLNA, HSP90AA1, HSPA1B, IGF1R, LDHB, MPO, PGK1, PPARG and YWHAG), two exosomal proteins (LGALS3BP and CFL1), and three proteins identified simultaneously as vesicular proteins and exosomal proteins (GAPDH, HSPA1A, and LDHA). Finally, 18 putative EVs proteins could be detected in the repertoire of candidate plasma tenderness proteins identify in this study.

QTL Investigation
As seen previously, 14 proteins were identified as located in cattle meat QTL for Shear force and/or Tenderness score from the network analysis (Table 2) (Table 2).

Discussion
As a potential rich source of biomarkers, secreted proteins are targeted by biologists for the discovery of biomarkers [65] especially because they reflect various states of the cells at real time under given conditions. More specifically, secreted proteins in plasma are promising for the identification of low invasive biomarkers circulating in the bloodstream. Therefore, we assumed that in silico prediction of the secretome might help us discovering candidate biomarkers for beef tenderness in the plasma. As a first step in the biomarker identification workflow [66], we designed a study based on the review of the literature and the aggregation of molecular data related to meat tenderness. According to Bonnet et al. [67], we performed a computational reconstruction of the secretome putatively linked to tenderness from the aggregated data, and searched for proteins secreted in the plasma. With this approach, we proposed a list of 71 putative plasma proteins to be investigated further as candidate plasma biomarkers for meat tenderness. Four other plasma candidates from recent literature will thereby expand this list through this discussion. Thus, from this final list of 75 candidate biomarkers, we propose a list of 33 proteins, which are particularly promising for meat tenderness (Table 6).
We report the 33 promising plasma candidate biomarkers for meat tenderness identified in this study. In brackets in the QTL column: chromosome associated with the Tenderness score and/or Shear force QTL. The first 29 promising candidates were selected when located in tenderness QTL (n = 24) and/or identified (n = 5) in [67]. The four plasma proteins reported at the bottom of table were obtained by overlapping between the BPA and the list of 67 putative muscle biomarkers published in [67]. These four proteins were predicted as secreted proteins (conventional pathways) using ProteINSIDE. "X" means that the protein was found in the Picard and Gagaoua 2019 and/or identify as promising candidate biomarkers.

Relevance of the Aggregated Dataset
Over the last two decades, 44 studies meeting our criteria of inclusion have identified genetic markers, and proteins or transcripts of which the abundance was related to tenderness. Some of them were proposed as muscle-derived biomarkers for meat quality [68]. These studies corresponded to less than 10% of the curated articles on meat tenderness. From this corpus, we aggregated a full compiled Atlas comprising 469 unique Gene Names, which we considered sufficient for further information mining. From this non-exhaustive dataset, we were able to identify 71 plasma candidate biomarkers for beef tenderness. Moreover, by comparison of the full compiled Atlas with the 67 proteins proposed recently in Picard et al. [68], four additional proteins (COL4A1, HSPA5, ORM1, PDIA3), both predicted as secreted proteins (with Signal-p and no TM) and found in the BPA, were included in our list of candidate biomarkers for meat beef tenderness. Thus, these results allowed to enrich, to 75 candidate plasma proteins, the list of candidates proposed in this study. The relevance of the list is supported by the good overview of tenderness mechanisms permitted by the data, as illustrated by GO term enrichment and their semantic analysis. The main pathways involved in meat tenderness (reviewed in [21,33]) were detected with our dataset as illustrated by the top 50 BP terms retrieved by a GO analysis (Table 3). Indeed, we report Biological Processes related to muscle structure and contraction (protein stabilization, muscle contraction and development, chaperone-mediated protein complex assembly, cell adhesion), muscle energy metabolism (gluconeogenesis, glycolytic process, oxidation-reduction process, carbohydrate metabolism), "post-mortem proteolysis" (aging, apoptotic process), "oxidative stress and HSP proteins" (cell detoxification, response to hydrogen peroxide, response to oxidative stress), and "metabolism, transport and cell signalling" (protein folding). The validation of the relevance of the aggregated dataset was a critical step prior to further computational analysis.

Reconstruction of the Secretome Linked to Tenderness and Identification of Secreted Proteins in Plasma
We propose for the first time a repertoire of secreted proteins related to tenderness. As predicted by bioinformatics, these proteins could be secreted through different pathways.

Proteins Predicted to Be Secreted through Conventional and Unconventional Pathways of Secretion (UPS)
From the aggregated dataset, 11.5 % of the proteins were predicted as secreted proteins through conventional-and 7.7% through alternative pathways. This is consistent with the report that 10-15 % of the human proteome is likely to be secreted through conventional and UPS secretory pathways [69,70]. However, although the bioinformatics reconstruction of the secretome with ProteINSIDE could identify secreted protein thanks to prediction algorithms, it did not enable to distinguish between proteins secreted into the surrounded extracellular fluid and proteins secreted into the bloodstream [67] Noteworthy, by overlapping the repertoire of predicted secreted proteins with a curated non-exhaustive bovine plasma atlas, we depicted 24% of them as putative plasma proteins. This result fits with the report by [71] that 31% of the secreted proteins of the human proteome are found in the plasma. However, the lower proportion of the secreted proteins in plasma in our dataset may be explained by the fact that our plasma atlas was very less that the 10,000 human proteins detected in serum/plasma curated from >500 published studies [70]. This suggests that by using a more complete plasma bovine atlas, we would increase by many the repertoire of secreted proteins in plasma. The semantic analysis of the enriched GO Biological Process associated with the repertoire of predicted secreted proteins in plasma (Table 4) revealed 6 associated biological pathways, linked to "cell adhesion", "apoptotic process", "endocytosis", "response to oxidative stress", "hydrogen peroxide metabolism", and "lipid metabolism". The most canonical pathways associated with the repertoire of 44 proteins were "homeostasis", "signal transduction (receptor tyrosine kinase signaling, and NR1H2/H3 mediated signaling)", "immune system (neutrophil degranulation)" and "transport of small molecules (plasma lipoprotein assembly, remodeling, ABC transporter ion channel, mitochondrial calcium ion transport . . . ). These results are in accordance with the literature relating to mechanism involved in non-vesicular UPS secretion [72]; such as "ABC transporter" reported as involved in the maintain of a stable physiological state and homeostasis in vertebrates [73]. Also, the liver X receptors LXR-α (NR1H3) and LXR-β (NR1H2), a subclass of nuclear receptors, were reported to bind the oxidized forms of cholesterol (or oxysterols), and activate the target gene expression [74]. These observations, suggest that lipid metabolism [75] and by consequence, in the light of our results, the secretion of proteins associated with lipid metabolism (conventional and UPS), could be involved in the tenderness. This is consistent with previous studies linking the lipid metabolism with the meat quality attributes flavour and tenderness [76,77].

Proteins Hypothetically Secreted in the Plasma
By overlapping the repertoire of proteins not hallmarked for secretion (i.e., without a signal P, Target P, or a GO term "secretion") with the bovine protein atlas, we retrieved proteins known to be found in the plasma. We therefore declared them as proteins hypothetically secreted in the plasma. The biological processes associated with these proteins were associated mainly with muscle contraction, protein stabilization, protein folding, chaperones, carbohydrate metabolism, and endocytosis. Moreover, six BP terms (four related to oxidant status, one to neutrophil degranulation and one to receptor-mediated endocytosis) were shared between the repertoire of secreted proteins in plasma and of hypothetically secreted proteins in plasma. While anti-oxidant proteins (PRDX6, MPO, and ATP2A2) were rather associated with the predicted proteins secreted the former, heat-shock proteins (HSPA1A, HSPA1B, HSP90AA1) were associated with the proteins hypothetically secreted in plasma. The most canonical pathways associated with the repertoire of 13 proteins included "cell-cell communication", "homeostasis", "muscle contraction", "metabolism of proteins", "metabolism of lipids (citric acid cycle and carbohydrate metabolism)", "programmed cell death", "cellular responses to external stimuli", "organelle biogenesis and maintenance (cilium assembly)", "autophagy", "extracellular matrix organization". Interestingly the primary cilia were described as involved in various pathways related to development and tissue homeostasis, such as Wnt [78] or Hedgehog [79] pathways. The muscle stem cells need a primary cilium for effective muscle regeneration [80]. The primary cilia were also reported as involved in other vesicular UPS [81].

Extracellular Vesicle Proteins as a Sub Repertoire of Tenderness Proteins Secreted in Plasma
During the last decade, extracellular vesicles (EVs) released by the cells have been described as key actors in intercellular communication in physiological conditions (e.g., heart and muscle development, angiogenesis) [82,83] and in pathogenesis especially in cancer [84]. The EVs are lipid bilayer particles composed of a range of different lipids and proteins (especially phospholipids, cholesterol and tetraspanin proteins), that can carry proteins, RNA and DNA in their aqueous core. EVs include microvesicles (MVs; 100-1000 nm size) or exosomes (30-100 nm size) and apoptotic bodies (1-5 µm) transporting proteins, mRNA, miRNA and lipids in the extracellular medium of cells and putatively in plasma because according to [85,86] all the bio-fluids (e.g., blood, urine, salive, lymphe, milk) contain EVs. Extracellular vesicles represent a potential source for biomarker discovery and can be used for drug and vaccine delivery conditions [87]. EVs are be considered as integrators of tissue physiology and whole-body homeostasis [88,89] EVs secretion is induced in response to extracellular signals such as ATP, interleukins, depolarization, thrombin receptor activation or by cell stress [90,91] Exosome secretion meanwhile can be induced by stress condition, micronutrient starvation, infection or cancer [92]. Recent studies have shown that skeletal muscle is also able to release EVs into the extracellular space [93,94] and to crosstalk with tissues and organs through this mechanism. In this study, we looked whether the hypothetically secreted proteins in plasma could be mapped to EVs. Supporting this hypothesis, we found that 36 % of the proteins were found in an atlas of vesicular proteins and 11 % in the exosome atlas. Therefore, we propose for the first time that EVs and exosome may be a possible reservoir of biomarkers for tenderness. We have identified 13 EVs proteins and two exosomal proteins in the dataset of hypothetically secreted proteins in plasma. Unexpectedly, we also found three vesicular proteins and two exosomal proteins (including the GAPDH protein in common) in the dataset of conventionally and unconventionally secreted proteins in plasma. Similarly, [70] also reported that proteins containing signal peptides that are secreted by the ER-Golgi pathway are also detected in extracellular vesicles. They suggested an unknown mechanism of sorting secreted proteins into these vesicles. Chauhan et al. [95] showed that the GAPDH protein is trafficked to the plasma membrane to be released in the extracellular matrix without use of the classic endoplasmic-Golgi secretion pathway but exosomes and secretory lysosomes.
To our knowledge, the association of EVs or exosomes with tenderness has never been reported. The biological significance of EVs tenderness proteins is unknown but their circulating level in the bloodstream could be a signature of the meat potential of the animals. Regarding their role, recent studies have suggested a role for EVs for the sharing of metabolites and other material between cells or tissues. According to Stahl et al. [96], EVs could operate as "independent metabolic units" that shuttle important molecules (enzymes, metabolites) for muscle homeostasis. Thus, we cannot exclude a role for EVs in unfavorable conditions especially following death of the animal (anoxia, pH and calcium release. The acid environment in muscle fibres after the animal death could promote the release of exosomes by muscle cells [97]. By delivering enzymes and/or metabolites involved in the glycolytic metabolism (e.g., LDHB and PGK1) to muscle cells post-mortem, the exosomes could compensate the early stop of glycolytic flux (glycolysis) independently of glycogen availability. EVs could also modulate the redox metabolism (myeloperoxidase (MPO), Thioredoxin-dependent peroxide reductase (PRDX3)) or address some HSP to delivery sites where they could play a crucial role in protecting the cells following death. Indeed, some HSP proteins such as the HSP70 [98,99] were reported in association with the membranes of EVs. More specifically, the HSP90 protein has been described as being exported via exosome vesicles [100,101].
Following network analysis we could include five new proteins found in plasma (CFL1, GC, PLEC, SLC4A1 and VCL) in the repertoire of tenderness hypothetically secreted proteins in plasma. These proteins have not been linked to tenderness so far, but at the exception of GC (vitamin D binding protein), they can be related to known pathways important for meat tenderization. The Cofilin 1, non-muscle (CFL1) is known to be involved in promoting actin polymerisation and organisation of actin filament, lipid metabolism, gene regulation and apoptosis [102]. This protein was also reported as associated with muscle lipid composition [103] Jia et al. [104] compared the post-mortem evolution of the proteome muscles differing in their tenderness (the Longissimus thoracis (tender) muscle and Semitendinosus (tougher) muscle). They reported a decrease in the levels of CFL1. The plectin (PLEC) and the vinculin (VCL) are two major structural components of the muscle cytoskeleton [105] located at the Z-discs [106]. These proteins are important proteins found in the costamere (which attaches myofibrils to the sarcolemma) that are essential for muscle fibre integrity and function (reviewed in [107]). Their proteolytic degradation post mortem leads to the disruption of the myofibrillar structure and to tenderisation of the meat. The SLC4A1 gene encodes the Cl − /HCO3 − anion exchanger 1, an acid loader that exchange one Cl − into cells for onw HCO 3 − out of cells, and thus is involved in the regulation of intracellular pH, especially in erythrocytes and kidney cells [108].

Review of the Literature
A computational workflow was created (Figure 1) to retrieve the data and aggregate them from available publications reporting meat tenderness. Briefly, we collected publications on meat tenderness by literature boolean operators: "meat OR beef AND tenderness AND biomarkers", "meat AND quality" and "muscle AND beef AND proteome (or "transcriptome", or "genetics") using MEDLINE (PubMed, https://www.ncbi.nlm.nih.gov/pubmed/), GOOGLE (Google Scholar, https://scholar.google.fr/) and CLAVIRATE (Web Of Science, https://clarivate.com/products/web-of-science/) analytics search until January 2018.

Parameters of Inclusion
All of the articles related to cattle meat tenderness were reviewed and curated based on the relevance and significance of the results. Only, molecular data related to the meat tenderness of Bos taurus and Bos indicus were conserved. Protein data could come from individual data. Only data with significant correlation of genetic polymorphism with tenderness, or differential abundances of transcripts or proteins according to tenderness as declared by the authors, were kept to build a meat tenderness aggregated dataset. A study associated with one GEO Dataset reporting transcriptomic data was analyzed with GEO2R (https://www.ncbi.nlm.nih.gov/geo/geo2r/) that enabled to compare two groups of samples according to tenderness. The differentially abundant transcripts between tenderness groups were included in our study.

Data Extraction
The molecular data collected from proteomic, transcriptomic or genetic studies were extracted from the articles and aggregated as follows. The proteins identifiers (ID) or gene symbols were retrieved from tables in Portable Document Format (PDF) or from supplementary data files of the publications. Data were extracted with Tabula (www.tabula.technology, Last update 11 February 2017).

Protein Identifiers Standardization
Protein ID and gene symbols were converted into the corresponding Gene Name identifiers (GN), as unique identifiers by use of three tools: Retrieve/ID Mapping tool of the Uniprot database (The UniProt 24), the Protein Identifier Cross-Reference service 25 and/or the ProteCONVERT tool of the ProteINSIDE web interface 26. Last conversion from ID to GN in February 2018.

Gene Ontology
In order to identify biological pathways associated with the aggregated dataset, Gene Ontology (GO) analysis was performed with the ProteINSIDE webservice (http://www.proteinside.org) [109] The GO enrichment analyses were achieved in the Human species in order to extend and promote GO interpretations because the bovine annotations remain limited. Only the Biological Process (BP) were considered. The Benjamini Hochberg (BH) adjusted P-values were considered to establish lists of significant enriched pathways in each dataset as compared to the whole genome. The GO_BP overview was carried out only with annotations with p-values < 0.001, minimum of annotated proteins ≥ 2. A table of the GO_BP overview was constructed in a semantic SimRel similarity-based Scatterplots with p-values associated to GO terms using REVIGO web tool (http://revigo.irb.hr/) [110] A visualization of the canonical pathways associated with the lists of candidate plasma proteins identified in the study was performed using Reactome tools (https://reactome.org/; voronoi hierarchical representation).

Prediction of the Secreted Proteins
In order to identify putatively secreted proteins belonging to the aggregated dataset, we used ProteINSIDE, a free web tool (http://www.proteinside.org) [109] that enables retrieving biological information from public databases in a single query. The secretion prediction module of ProteINSIDE runs a local version of SignalP 4.1. From the sequences of input ID proteins, it looks for signal peptide type sequences. The program also checks if proteins are related to a secretory function by looking for GO secretion annotation terms. The aggregated dataset was submitted to a computational prediction of proteins secreted using "custom analysis", "bovine species", "signal P" and "increase cleavage site sensitivity (D-cutoff 0.34)" parameters (version of Database 1.2.11, CBS signal-P 4.1 software, May 2018). To declare proteins as "predicted secreted proteins", we used the following criteria.
(1) File tab "Secreted Protein", Signal-P score > 0.5 and Target-P score ≤ 2 to identify the proteins predicted as secreted through a signal-P sequence and/or a transmembrane domain (TM) (named "conventional predicted secreted proteins"). (2) File tab "other secreted protein", Target-P score ≤ 3 with GO term associated to identify the proteins predicted as secreted through an unconventional pathway of secretion (or UPS) without signal-P (named "UPS predicted secreted proteins") [72,111]. The conventional-and UPS-predicted secreted proteins were merged in a single repertoire referred to as predicted secreted proteins. All of the proteins not identified as predicted secreted proteins were "the remaining proteins" (aggregated data minus secreted proteins).

Prediction of Plasma Location
In order to search for the proteins that may be found in the plasma, we compared protein lists using VIB / UGent (http://bioinformatics.psb.ugent.be/webtools/Venn/). The comparisons were performed between the repertoire of predicted secreted proteins and a "Bovine Plasma proteome Atlas" (BPA, n = 1101 plasma proteins, which were merged from publications [67] and experimental data (Supplementary Data 3). Similarly, a comparison between the remaining proteins and the BPA was performed to detect hypothetically secreted proteins and found in plasma.

Network Analysis and Protein-Protein Interactions
In order to enrich the list of putative plasma proteins, we used the academic Cytoscape open source software ® (Version 3.7.2, https://cytoscape.org/) [112] with the Psicquic plugging web service (https://apps.cytoscape.org/apps/psicquicuniversalclient, up to date, 2017-12-17) [113]. The parameters for network analysis were "MINT database", "human species". The proteins that interact with proteins within our dataset were named "interactors". For representation, the 13 predicted secreted proteins in plasma (conventional and alternative pathways) are shown in purple ellipses. The 44 hypothetically secreted proteins in plasma are shown in pink ellipses. The green rectangle refer to interactor identify using the MINT Cytoscape analysis.

Search for QTL
By using the ProteQTL module of ProteINSIDE, we searched for the location of genes encoding the proteins of interest within published Quantitative trait loci (QTL) for tenderness.

Identification of the EVs Proteins
In order to test the hypothesis that membrane-derived vesicles secretion could be associated with tenderness, we compared the repertoires of candidate proteins with the Human Protein Atlas (HPA) that lists the vesicular proteins experimentally detected in the vesicles (referred to as "vesicular protein Atlas", n = 1998; 2019 October, 28th; https://www.proteinatlas.org/) and the Exosome protein atlas (n = 100, 25 October 2019, http://exocarta.org/) that lists the proteins detected in exosomes.

Dataset Descriptors
Four datasets were generated in this study. The aggregated dataset merged from three individual lists related to beef tenderness, namely a "proteomic dataset", a "transcriptomic dataset", and a "genetic dataset" The aggregated dataset (named "full compiled atlas") was deposited as ".xls" files at the French INRA public repository (Portail Data INRA, data.inra.fr) hosted by Dataverse.org and is directly available at [63]. In addition, the two repertoires generated by reconstruction of the secretome were named the "predicted secreted proteins in plasma" and the "proteins hypothetically secreted in plasma". Finally, the merged of the "predicted secreted proteins in plasma", the "proteins hypothetically secreted in plasma" and the "interactors" putatively found in the plasma" generated the final list of candidate plasma proteins proposed by this study as putative low-invasive candidates for meat tenderness in beef cattle.

Conclusions
This study is the first to use data aggregated from a corpus of published data for the purpose of identifying novel meat tenderness in muscle (thanks to PPi) and in the plasma. We propose for the first time a non-exhaustive list of 75 candidate biomarkers for tenderness in the plasma. Combined with QTL data and recent literature, 33 are of particular interest for further evaluation and validation for future low-invasive approach, among which four proteins recently reported as muscle tenderness biomarkers and found in plasma. Another original finding of this study is that the secretion pathway of 13 of these plasma proteins could be the membrane-derived vesicle secretion. The 33 plasma candidate biomarkers for meat tenderness identified in this study require further assessment and validation.
Supplementary Materials: Supplementary materials can be found at http://www.mdpi.com/1422-0067/21/2/664/s1; Supplementary Data 1. Reactome representation of the canonical pathways associated with the 13 "predicted secreted proteins in plasma" identified in this study; Supplementary Data 2. Reactome representation of the canonical pathways associated with the 44 "hypothetically secreted proteins in plasma" identified in this study; Supplementary Data 3. List of the 1101 Gene Names used as Bovine Proteome Atlas (BPA).