Introduction
Plants are the main contributors to the production of biomass including carbohydrates, proteins, lipids, cellulose and other biomaterials. Plant proteins including enzymes, regulatory and structural proteins play important biological roles in regulating plant growth and development. Plant proteins are synthesized within a cell and transported to different subcellular locations including extracellular space or matrix to perform their biological functions. This process often is called protein sorting or targeting (Foresti and Denecke, 2008; Rose and Lee, 2010). Plant cells contain a cell wall, a plasma membrane, choloroplasts, mitochondria, a large vacuole, a nucleus, endoplasmic reticulum (ER), a Golgi apparatus, peroxisomes, cytosol, etc. Membrane proteins can be embedded or attached to plasma membrane, organelle membrane or endomembrane systems.

Identification and analysis of protein subcellular locations in eukaryotes is one of the important subjects for annotating a proteome. In a plant species, proteins secreted to the extracellular space or matrix, which includes the cell wall, are collectively called a “secretome” (Agrawal et al., 2010; Lum and Min, 2011a). The term secretome was first introduced by Tjalsma et al. (2000) to denote the complete set of proteins in Bacillus subtilis processed by the secretory pathway, which included protein secreted to the extracellular space and also proteins involved in the pathway. However, recently it was more often limited, as in this work, to represent only the secreted, extracellular portion - including cell wall proteins - of the proteome (e.g., Greenbaum et al., 2001; Hathout, 2007; Bouws et al., 2008; Agrawal et al., 2010; Lum and Min, 2011b). A plant secretome consists of primarily cell wall proteins, proteins involved in cell wall metabolism, and extracellular enzymes and signal molecules involved in defense of pathogens (Isaacson and Rose, 2006; Kamoun, 2009; Lum and Min, 2011a). Secreted enzymes, particularly hydrolases such as α-amylase and α-glucosidases, have been well studied using germinating barley seeds as a model system. These hydrolases were synthesized in the aleurone layer and secreted into the endosperm to break down starch and other storage reserves (Ranki and Sopanen, 1984; Jones and Robinson, 1989; Finnie et al., 2011 for review). Recently, advances in proteomic analytic techniques along with the complete sequencing of Arabidopsis thaliana and Oryza sativa genomes resulted in many secreted proteins, including the cell wall proteome, being identified (Boudart et al., 2007; Agrawal et al., 2010; Lum and Min, 2011a). These identified secreted proteins mainly consist of cell wall proteins in Arabidopsis (see Jamet et al., 2008 for review) and some enzymes such as GLP1 involved in pathogen defense (Oh et al., 2005). Using a leaf or seed cell suspension culture, secreted proteins were identified with 2D-gel electrophoresis coupled with liquid chromatography mass spectrometry analysis in rice, Medicago and sorghum (Jung et al., 2008; Kusumawati et al., 2008; Cho et al., 2009; Ngara and Ndimba, 2011). A large number of secreted proteins were also identified from root exudates using aseptically grown seedlings of rice and Arabidopsis (Shinano et al., 2011; De-la-Pena et al., 2010). Experimental systems, analytical techniques, and related bioinformatics tools used for plant secretome study were recently comprehensively reviewed (Agrawal et al., 2010; Meinken and Min, 2012; Alexandersson et al., 2013; Kraus et al., 2013; Caccia et al., 2013).

Classical eukaryotic secreted proteins contain a secretory signal peptide at the N-terminus that directs proteins to the rough ER for completing protein synthesis and then transports them to the Golgi complex for protein targeting (von Heijne, 1990). The signal peptide, typically 15~30 amino acids long, is often cleaved off during translocation across the endomembrane systems. Classical secreted proteins can be computationally predicted relatively accurately (Min, 2010). Recently we analyzed all manually curated and annotated secreted plant proteins in the UniProtKB/Swiss-Prot dataset and found 87% of them could be predicted to have a signal peptide by all three predictors used (Lum and Min, 2011a). The accuracy of secretome prediction could be further improved by using a new version of SignalP (SignalP 4.0) combined with other tools including TMHMM for identifying transmembrane proteins and PS-Scan for identifying ER luminal proteins (Min, 2010; Melhem et al., 2013).

With improvements in sequencing technology and the reduced cost of sequencing, the genomes of more and more plant species are being completely sequenced. Currently there are 32 land plants with complete or draft genome sequences available and 73 land plant species with genome sequencing in progress (http://www.ncbi.nlm.nih.gov/genomes/static/gpstat.html). There are also assembled expressed sequence tag (EST) data in plants available for identifying potential genes encoding secreted proteins in more than 200 species (PlantGDB, http://www.plantgdb.org/prj/ESTCluster/) (Duvick et al., 2008). As a result of genome sequencing, the number of protein sequences available is increasing rapidly.

In addition to the classical secreted proteins, a large number of leadless, non-classical, secreted proteins (LSP), i.e. not having a secretory signal peptide, have been identified in plants (Jung et al., 2008; Agrawal et al., 2010; Ding et al., 2012 for review). These proteins have not been curated in the UniProtKB. Therefore there is a need to have a central knowledgebase providing plant protein subcellular locations for the plant research community to access the available information and deposit experimental evidence for newly characterized proteins. In order to provide such a central plant secretome related resource portal, we developed the Plant Secretome and Subcellular Proteome KnowledgeBase (PlantSecKB) (http://proteomics.ysu.edu/secretomes/plant.html), which includes predicted and manually curated protein subcellular locations from plant proteomes as well as predicted proteins from EST data in plants. Though our focus is on plant secretomes, the information on proteins located in other subcellular locations is also provided. A tool for supporting community manual curation of plant protein subcellular locations can be accessed through the database interface.

1 Methods of Database Construction
1.1 Data collection
PlantSecKB was constructed primarily with the sequence data obtained from two sources: plant protein sequences extracted from UnitProtKB (2013-04 Release) (http://www.uniprot.org/) and protein sequences predicted from assembled EST data compiled by the PlantGDB project (http://www.plantgdb.org/prj/ESTCluster/). The proteins predicted from the recently sequenced sacred lotus (Nelumbo nucifera Gaertn.) genome were also integrated into this database (Ming et al., 2013; Lum et al., 2013). Protein sequences in the EST data were predicted using the OrfPredictor tool (http://proteomics.ysu.edu/tools/OrfPredictor.html) with BLASTX input against the UniProt/Swiss-Prot database, and TargetIdentifier (http://proteomics.ysu.edu/tools/TargetIdentifier.html) was used to examine if an EST was full-length (Min et al., 2005a, 2005b).

1.2 Computational methods for prediction of protein subcellular locations
The software tools used in this study include SignalP 3.0 and 4.0, TargetP, Phobius, WoLF PSORT, TMHMM, PS-Scan, and FragAnchor. The website links for these tools and related references can be found in our website (http://proteomics.ysu.edu/tools/subcell.html). Except FragAnchor, we used the standalone tools installed on a local Linux system for data processing. The commands for how to run them often can be found in the “readme” page in each downloaded package and were summarized by Lum and Min (2013). In brief, SignalP 4.0 was used for secretory signal peptide prediction (Petersen et al., 2011). However, we also included prediction information from SignalP 3.0 (Bendtsen et al., 2004b) as it provides more accurate cleavage site prediction than SignalP 4.0 (Petersen et al., 2011). Phobius is a combined signal peptide and a transmembrane topology predictor (Käll et al., 2007). TargetP predicts the presence of any signal sequences such as signal peptide (SP), chloroplast transit peptide (cTP) or mitochondrial targeting peptide (mTP) in the N-terminus (Emanuelsson et al., 2000; Emanuelsson et al., 2007). TMHMM uses a hidden Markov model (HMM) to predict the presence and topology of transmembrane helices and their orientation to the membrane (in/out) (Krogh et al., 2001). PS-Scan was used to scan the PROSITE database (http://www.expasy.org/tools/scanprosite/) for removing ER targeting proteins (Prosite: PS00014) (de Castro et al., 2006; Sigrist et al., 2010). FragAnchor was used to identify the glycosylphosphatidyinositol (GPI) anchored proteins (GAP) from the proteins which were predicted as containing a signal peptide by SignalP 4.0 (Poisson et al., 2007). WoLF PSORT predicts multiple subcellular locations including choloroplast, cytosol, cytoskeleton, ER, extracellular (secreted), Golgi apparatus, lysosome, mitochondria, nuclear, peroxisome, plasma membrane, and vacuolar membrane (Horton et al., 2007). The default parameters for eukaryotes or plants, if available, were used for all the programs. Our previous evaluation found that including WoLF PSORT for plant secretome prediction resulted in an accuracy decrease due to a significant decrease in the prediction sensitivity (Min, 2010). Thus, it was not used for secretome prediction but only for prediction of some other subcellular locations.

For the assignment of a subcellular location of a protein, the UniProtKB annotated subcellular location and our manual curation take precedence over computational prediction. Thus, only proteins not having an annotated subcellular location are subjected to computational assignment of their subcellular locations. The information produced by all the tools, however, is available for all plant proteins. Some of the proteins may have more than one subcellular location. The following criteria are applied for computational classification of protein subcellular locations:

Membrane proteins: A protein predicted to contain one or more transmembrane domains by TMHMM is classified as a membrane protein. However, if there is only one transmembrane domain predicted and that is located within the N-terminus 70 amino acids, and also a signal peptide is predicted by SignalP 4.0, this protein is not counted as a membrane protein.

Chloroplast proteins: A protein predicted as “C” (for chloroplast) for subcellular location by TargetP is classified as a chloroplast protein. If it is also classified as a membrane protein, then it is further classified as chloroplast membrane protein.

Mitochondrial proteins: A protein predicted as “M” (for mitochondrial) for subcellular location by TargetP is classified as a mitochondrial protein. If it is also classified as a membrane protein, then it is further classified as mitochondrial membrane protein.

ER proteins: Proteins predicted to contain a signal peptide by SignalP 4.0 and an ER target signal (Prosite: PS00014) by PS-Scan were treated as luminal ER proteins.

Complete secretomes: A secretome is all secreted proteins from a species. Only proteins that are predicted to have a secretory signal peptide by all three predictors - SignalP 4.0, Phobius, and TargetP - and that are not classified as any of the above categories are included in the secreteome. However, proteins that are not classified as any of the above categories and are predicted to have a signal peptide by one or two of the predictors are assigned as “weakly likely secreted” or “likely secreted” as our previous evaluation revealed that a signal peptide in some annotated secreted proteins can only be detected by one or two predictors (Lum and Min, 2011a). Using all three predictors, which increases the specificity of secretome prediction, improves prediction accuracy (Min, 2010; Melhem et al., 2013). All manually curated secreted and extracellular proteins are included in the complete secretomes.

Curated secreted proteins: This category includes proteins which are annotated to be “secreted” or “extracellular” or "cell wall" in the subcellular location from the UniProtKB/Swiss-Prot data set which are “reviewed”. It also includes manually collected secreted proteins from recent literature by our curators.

GPI-anchored proteins: Signal peptide containing proteins that were predicted to have a GPI anchor by FragAnchor were further classified as GPI-anchored proteins. Protein sequences predicted to have a signal peptide and a GPI anchor may attach to the outer leaflet of the plasma membrane or be secreted becoming components of the cell wall. These proteins are involved in signaling, adhesion, stress response, and cell wall remodeling or play other roles in growth and development (Borner et al., 2002; Borner et al., 2003; Gillmor et al., 2005; Simpson et al., 2009).

Proteins in other subcellular locations: Other subcellular locations including cytosol (cytoplasm), cytoskeleton, Golgi apparatus, lysosome, nucleus, peroxisome, plasma membrane and vacuole were predicted by WoLF PSORT.

1.3 Computational prediction accuracies of protein subcellular locations
The prediction methods we used above were developed based on our previous evaluation of computational tools (Min, 2010; Meinken and Min, 2012; Melhem et al., 2013). To estimate the prediction accuracies of our methods for each subcellular location we used two datasets (Table 1). Dataset A consists of 15 028 proteins. This dataset contains proteins from the UniProtKB/Swiss-Prot dataset with a curated subcellular location. Proteins having multiple subcellular locations or labeled as “fragment” were excluded. Dataset B consist of 6 908 proteins which were generated from Dataset A after excluding entries having a term of “by similarity” or “probable” or “predicted” in subcellular location annotation. In comparing with other methods using a single tool, our method - i.e. using a combination of multiple tools including SignalP 4.0, TargetP, and Phobius for secretory signal peptide prediction and PS-Scan for removing ER proteins and TMHMM for removing membrane proteins - significantly improved the prediction accuracy for secretomes (Min, 2010; Meinken and Min, 2012). For secretome prediction our method had reached a sensitivity of 91.1%, a specificity of 98.7%, and a Mathews’ correlation coefficient (MCC) of 88.5% for dataset A; and a sensitivity of 76.8%, a specificity of 98.9%, and a MCC of 74.5% for dataset B, which were much better than using WoLF PSORT or MultiLoc alone (Meinken and Min, 2012). Thus the prediction of secreted proteins is relatively reliable. The accuracies for predicting other subcellular locations still need to be improved.

Table 1 Evaluation of prediction accuracies of plant protein subcellular locations

1.4 Manual curation and community annotation
PlantSecKB supports community curation of subcellular locations of plant proteins based on published experimental evidence. A submission tool was developed for the community to provide subcellular location annotation of a protein and a literature source to support its annotation. After our curator’s validation, these data are also incorporated into the database. Currently, based on published experimental evidence, we have manually curated 736 total secreted proteins from rice (Jung et al., 2008; Cho et al., 2009; Cho and Kim, 2009; Chen et la., 2009; Zhang et al., 2009; Shinano et al., 2011), Arabidopsis (De-la-Pena et al., 2010), and sorghum (Ngara et al., 2011). Manual curation is an ongoing process, thus more secreted proteins will be manually curated and integrated into the database in the future from the community and our curators. The information from computational prediction, UniProtKB annotation and manual curation is integrated and displayed on the annotation page (Figure 1). The annotated entries are linked to the tools used, UniProtKB, the RefSeq database and PubMed in the National Center for Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov/). 

Figure 1 Overview of the PlantSecKB user interface and annotation page

 
2 Overview of the Database Content and Tools
2.1 Data and tool access
The PlantSecKB is accessed through the database web interface at http://proteomics.ysu.edu/secretomes/plant.php. The interface provides various utilities for searching proteins obtained from UnitProtKB, links to BLAST, an EST data search page, and the community annotation page (Figure 1). All plant proteins obtained from UniProt can be searched using UniProt accession number (AC) or ID, gene name, key word(s) in protein function or species. Sub-proteomes including curated secreted proteins, complete secretome, mitochondrial membrane proteins, ER proteins, and others can be searched or downloaded by selecting species from a species list for those having greater than 1 000 protein sequences. Species having fewer than 1 000 protein entries can be searched by inputting a species name. The BLAST utility can be accessed through a link on the interface for searching all plant proteins or secretomes. The interface also provides a link to an EST data search page. EST data can be searched using EST identifier, keyword(s), species or BLAST.

The annotation display page for each UniProt protein contains information obtained from the following three sources: (1) the features predicted using computational approaches using the seven programs mentioned above; (2) subcellular locations annotated in UniProtKB; and (3) our manual curation with experimental evidence obtained from recent literature. The overview of the database features is shown in Figure 1. Manually curated secreted proteins consist of proteins retrieved from UniProtKB/Swiss-Prot with subcellular locations labeled as “reviewed”, as well as proteins curated by our curators. The curated proteins from internal curation and the community are supported with experimental evidence for their subcellular location annotation and related literature. The annotation page also contains the primary protein sequence (Figure 1).

EST data annotation contains the primary EST sequence, predicted protein peptide sequence using OrfPredictor (Min et al., 2005a), functional annotation based on BLASTX, prediction of completeness of the open reading frame using TargetIdentifier (Min et al., 2005b), and related information generated with the tools for subcellular location prediction based on predicted protein sequences. As EST data may contain errors introduced in sequencing and assembling, caution needs to be taken when using the data. Nevertheless, EST information provided in the database will be useful for data mining and designing experiments for further examining the gene function and subcellular locations of encoded proteins.

2.2 Data summary
PlantSecKB contains a total of 1 415 921 protein sequences including 33 643 entries from the UniProt/Swiss-Prot dataset (curated and reviewed) and 1 355 593 from UniProt-TrEMBL (unreviewed) with an additional 26 685 proteins predicted from the newly sequenced genome of sacred lotus (Ming et al., 2013; Lum et al., 2013). The main categories of subcellular proteomes for species having more than 7 000 entries are summarized in Table 1. Curated secreted proteins, ER proteins and lysosome proteins are not listed in Table 1. There were only 7 lysosome proteins in A. thaliana identified and no lysosome proteins were predicted in other species. There are a total of 2 774 curated secreted proteins, which are mainly obtained from A. thaliana and O. sativa subsp. japonica with 1 247 and 559 entries, respectively. It should be noted that the number of total protein entries in a species is the number collected in the UniProtKB, which can be greater than a complete or reference genome, as there are some redundancies or duplicates in some protein entries. For example, O. sativa subspecies japonica has 99 984 entries in PlantSecKB and only 63 544 entries in its complete proteome set, and A. thaliana has 53 847 entries in PlantSecKB and only 31 908 entries in the complete proteome set in UniProtKB (http://www.uniprot.org/taxonomy/complete-proteomes).

An overall trend observed is that plants with relatively small proteome sizes have a relatively small number and a relatively lower proportion of secreted proteins, such as in single-celled green algae. For example, Osterococcus species has less than 100 secreted proteins predicted (1.2%), and moss (Physcomitrella patens) has 781 secreted proteins predicted (2.9%) (Table 2). On average the secretome accounts for about 4.0%~7.5% of the proteome in monocot and dicot plants based on our prediction estimations. The secretome percentages reported in this study are slightly lower than we reported previously. This is due to the fact that our previous study used SignalP 3.0, whereas this study used SignalP 4.0 which has a higher specificity (Lum et al., 2013; Petersen et al., 2011).

Table 2 Summary of subcellular proteomes in different plant species in PlantSecKB

 
The average predicted proteome sizes and distributions of subcellular proteomes are summarized in Table 3 using 9 species or subspecies in each category of green algae, monocot and dicot plants listed in Table 2. Lotus japonicus, a dicot, was the only species not used for this analysis due to incompleteness of its proteome. The average predicted proteome size is much smaller in green algae, thus each subcellular proteome consists of a smaller number of proteins (Table 3). Comparing monocots and dicots, the distribution percentages of secreted proteins, chloroplast membrane proteins, vacuolar proteins, and plasma membrane proteins were not significantly different. However, monocots had a significantly higher proportion of proteins predicted as mitochondria (both membrane and non-membrane) and chloroplast membrane, and dicots had significantly more proteins predicted as cytosol and nucleus (Table 3). Whether these observed differences in subcellular proteome distributions between monocots and dicots are caused by computational tools or are real with biological or evolutionary significances needs further investigation.

Table 3 Comparison of subcellular proteome distribution in green algae, monocot and dicot plants

 
3 Comparative Analysis of Secretomes
Complete comparative evolutionary analyses of plant secretomes or other sub-proteomes were beyond the scope of this study. However, as complete secretome or other sub-proteome sequences can be downloaded directly from our database, it would facilitate further detailed comparative study of these sub-proteomes in different species. As an example, we performed a comparative analysis of secretomes using a set of representative plants including three monocots (Brachypodium distachyon, Oryza sativa subsp. japonica, Zea mays), three dicots (Arabidopsis thaliana, Populus trichocarpa, Solanum lycopersicum), and two mosses (Physcomitrella patens subsp. patens, Selaginella moellendorffii) (Table 4 and Table 5). We used the blastclust tool in the BLAST package with a cutoff of 95% identities in the aligned pair to remove or reduce redundancy. Thus non- or less redundant secreteomes were used for comparisons. To provide an overview of the functionalities of secretomes in plants, we carried out Gene Ontology (GO) analysis of representative secretomes of the 8 selected plant species. The secretomes were used to search the UniProt/Swiss-Prot dataset with BLASTP with a cutoff E-value of 1e-10. GO information was retrieved from UniProt ID mapping data (http://www.uniprot.org/downloads) and analyzed using GO SlimViewer with plant specific GO terms (McCarthy et al., 2006). Comparison of GO biological process and molecular function classification of secretomes of the selected species was summarized in Table 4. Plant secreted proteins are involved in more than 40 different biological processes including metabolic and catabolic processes, response to stress and biotic or abiotic stimulus, carbohydrate, lipid and protein metabolic processes, multicellular organismal development, etc. Molecular function classification revealed that plant secretomes consist of a large number of hydrolases (~30%) and tranferases (7%~9%), and that a large proportion of them have various binding activity (~40%) or catalytic activity (12%~15%). It should be noted that GO classification was only an estimate of the distributions of each category as many secreted proteins have not been classified in GO.

Table 4 Gene Ontology classification of secreted proteins in different plant species

Table 5 Comparison of protein families in secretomes of representative plant species

 
The functionalities of secreted proteins were further analyzed using rpsBLAST to search against Pfam in the Conserved Domain Database (CDD) (Marchler-Bauer et al., 2009). The results of Pfam analysis for a species having 20 or more members in a Pfam were summarized in Table 5. A complete list of Pfams can be found in Supplementary Table 1. The detailed analysis of molecular functions in secretomes searching Pfam revealed the difference in protein families among different species, including both variations in the number of members in a given Pfam and species specific Pfams (Table 5). Noticeably there were twice as many secreted peroxidase proteins in rice compared to Arabidopsis (Table 5). Plant peroxidases have multiple tissue-specific functions e.g., removal of hydrogen peroxide from chloroplasts and cytosol, oxidation of toxic compounds, biosynthesis of the cell wall, and defense responses towards wounding (Sottomayor and Barceló, 2004). The glycosyl hydrolases are suggested to have valuable applications in modifying plant cell wall architecture and in the development and characterization of new bioenergy and feedstocks (Lopez-Casado et al., 2008). The rice secretome consists of 31 members of Glyco-hydro-18 (GH18) and 26 of GH32N while only two GH18 and 6 GH32N were identified in the Arabidopsis secretome. We also observed a number of Pfams having more members in rice than in other species. These Pfams include dirigen-like protein, multicopper oxidase, pollen allergen, cytochrome P450, etc (Table 5). It should be noted that these predicted secreted cytochrome P450 proteins most likely are false positives as there is no secreted cytochrome P450 protein reported with experimental evidence in plants. Wen et al. (2007) reported a cytochrome P450 presented in the pea root cap secretome. However, its presence might represent leakage that occurs during the cell separation process. In general, moss species have fewer secreted proteins as well as a smaller member number in a given Pfam due to their small genomes. However, we noted that the lycophyte model organism Selaginella moellendorffii has 20 members of D-mannose binding lectin family, while other plant species have less than 10 members in this Pfam, except Populus trichocarpa, which has 30 members. Species-specific secreted proteins are also observed, such as corn, which has 30 members of Zein seed storage protein and Physcomitrella patens (subsp. patens), which has 61 members of a protein with an unknown function (DUF4100).

4 Discussion
We constructed the PlantSecKB to provide a resource for the plant research community. As the subcellular location(s) of a given protein curated by UniProtKB or by us were considered first in assigning a subcellular location, these assignments are based on traceable literature with experimental evidence, and thus fairly reliable. However, the subcellular locations assigned based on the computational prediction will depend on the accuracy of the tools used. We have evaluated the prediction accuracy of the methods we used in this study and compared it with the accuracies of other methods (Table 1) (Min, 2010; Meinken and Min, 2012). We concluded the prediction of secreted proteins is relatively reliable. However, false positives and false negatives certainly exist. For example, a number of P450 enzymes were predicted to be secreted proteins, which are most likely false positives.

We also predicted other subcellular locations including mitochondrial, chloroplast, vacuole, nucleus, and others based on the predictions of TargetP and WoLF PSORT. Our evaluation on the prediction accuracies of these subcellular locations revealed that the accuracies of the tools we used, even though they are best among available tools, are still not satisfactory due to relatively low prediction sensitivities for these subcellular locations (Table 1) (Meinken and Min, 2013). With the exception of mitochondrial and cytosol proteins, however, the specificities for those subcellular locations including chloroplast, ER, Golgi apparatus, nucleus, plasma membrane, vacuole and cytoskeleton are acceptable (>89%). Thus, proteins predicted in those subcellular locations are relatively reliable, though they still need to be cautiously examined with experiments. Recently, several new tools were developed including the Cell-PLoc servers (Chou and Shen, 2008), MultiLoc2 (Blum et al., 2009), and others (Meinken and Min, 2012). These tools and their related publications can be found at our website (http://proteomics.ysu.edu/tools/subcell.html) (Meinken and Min, 2012). As standalone tools are not available for some of them, such as Cell-PLoc, or some standalone tools are too slow for processing a large data set, such as MutliLoc2, we were not able to use them for our data processing. However, we suggest users utilize these tools to get a second prediction for proteins of interest as our experience showed that using multiple tools improves prediction specificity.

Based on several recent large-scale secretome studies in plants, non-classical, i.e. leadless secretory proteins (LSPs) were observed to account for more than 50% of the total identified secretome, supporting the existence of novel secretory mechanisms independent of the classical ER-Golgi secretory pathway (Agrawal et al., 2010 for review; Jung et al., 2008; Cheng and Williamson, 2010; Ding et al., 2012). Mammalian and bacterial LSPs have been collected and used to implement the prediction software, SecretomeP, for predicting these proteins (http://www.cbs.dtu.dk/services/SecretomeP/) (Bendtsen et al., 2004a). Because the tool has not been trained with plant-specific data and the accuracy for predicting plant LSPs could not be evaluated, we did not include this tool in our data processing.

The PlantSecKB strives to serve as a portal for plant researchers to search plant protein subcellular locations with an emphasis on secreted proteins. The EST sub-database is expected to facilitate EST data mining for secreted proteins from expressed data, which is particularly useful for plant species not completely sequenced or having only a limited number of cDNA sequences. The collection and curation of secreted plant proteins, particularly LSPs, from literature with experimental evidence requires continuous efforts from the plant research community. We have implemented a curation tool accessible through PlantSecKB for the community to manually curate subcellular locations of plant proteins having experimental evidence. The utility described in PlantSecKB, together with our recently implemented Fungal Secretome KnowledgeBase (FunSecKB) (Lum and Min, 2011b), is anticipated to provide a search, download, and curation system that will help the plant community to further understand secretome biology. It can also be used to explore various potential applications and their interactions of plant and fungal secreted proteins for plant pathogen control and breeding for stress resistant varieties (Kim et al., 2009).

Authors' contributions
GL and JM implemented the database, JO and SF manually curated secreted proteins, XJM conceived of the study, designed the procedure of data processing. XJM, JM and GL analyzed the data and prepared the manuscript. All authors read and approved the final manuscript.

Acknowledgments
The work is funded by the Ohio Plant Biotechnology Consortium [grant 2011-001] (through the Ohio State University, Ohio Agricultural Research and Development Center), and Youngstown State University (YSU) Research Council [grant 2010-2011 and 12-11]. The work is also supported by a YSU Research Professorship and the College of Science, Technology, Engineering, and Mathematics Dean’s reassigned time for research to XJM. JM was supported with a graduate research assistantship by the Center for Applied Chemical Biology at YSU.

References
Agrawal G.K., Jwa N.S., Lebrun M.H., Job D., and Rakwal R., 2010, Plant secretome: unlocking secrets of the secreted proteins, Proteomics, 10: 799-827
http://dx.doi.org/10.1002/pmic.200900514 
 
Alexandersson E., Ali A., Resjö S., and Andreasson E., 2013, Plant secretome proteomics, Front. Plant Sci., 4: 9
http://dx.doi.org/10.3389/fpls.2013.00009

Bendtsen J.D., Jensen L.J., Blom N., von Heijne G., and Brunak S., 2004a, Feature based prediction of non-classical and leaderless protein secretion, Protein Eng. Des. Sel., 17: 349-356
http://dx.doi.org/10.1093/protein/gzh037 
 
Bendtsen J.D., Nielsen H., von Heijne G., and Brunak S., 2004b, Improved prediction of signal peptides: SignalP 3.0, J. Mol. Biol., 340: 783-795
http://dx.doi.org/10.1016/j.jmb.2004.05.028 
 
Blum T., Briesemeister S., and Kohlbacher O., 2009, MultiLoc2: integrating phylogeny and Gene Ontology terms improves subcellular protein localization prediction, BMC Bioinformatics, 10: 274
http://dx.doi.org/10.1186/1471-2105-10-274 
 
Borner G.H., Lilley K.S., Stevens T.J., and Dupree P., 2003, Identification of glycosylphosphatidylinositol-anchored proteins in Arabidopsis. A proteomic and genomic analysis, Plant Physiol., 132: 568-577
http://dx.doi.org/10.1104/pp.103.021170 
 
Borner G.H., Sherrier D.J., Stevens T.J., Arkin I.T., and Dupree P., 2002, Prediction of glycosylphosphatidylinositol-anchored proteins in Arabidopsis. A genomic analysis, Plant Physiol., 129: 486-499
http://dx.doi.org/10.1104/pp.010884 
 
Boudart G., Minic Z., Albenne C., Canut H., Jamet E., and Pont-Lezica R., 2007, Cell wall proteome, In: Samaj S., and Thelen J. (eds.), Plant Proteomics, Springer, pp.169-185

Bouws H., Wattenberg A., and Zorn H., 2008, Fungal secretomes-nature's toolbox for white biotechnology, Appl. Microbiol. Biotechnol., 80: 381-388
http://dx.doi.org/10.1007/s00253-008-1572-5 
 
Caccia D., Dugo M., Callari M., and Bongarzone I., 2013, Bioinformatics tools for secretome analysis, Biochim. Biophys. Acta., S1570-9639

Chen X.Y., Kim S.T., Cho W.K., Rim Y., Kim S., Kim S.W., Kang K.Y., Park Z.Y., and Kim J.Y., 2009, Proteomics of weakly bound cell wall proteins in rice calli, J. Plant Physiol., 166: 675-685
http://dx.doi.org/10.1016/j.jplph.2008.09.010 
 
Cheng F.Y., and Williamson J.D., 2010, Is there leaderless protein secretion in plants? Plant Signal Behav., 5: 129-131
http://dx.doi.org/10.4161/psb.5.2.10304  

Cho W.K., and Kim J.Y., 2009, Integrated analyses of the rice secretome, Plant Signal Behav., 4: 345-347
http://dx.doi.org/10.4161/psb.4.4.8198

Cho W.K., Chen X.Y., Chu H., Rim Y., Kim S., Kim S.T., Kim S.W., Park Z.Y., and Kim J.Y., 2009, Proteomic analysis of the secretome of rice calli, Physiol. Plant, 135: 331-341
http://dx.doi.org/10.1111/j.1399-3054.2008.01198.x 
 
Chou K.C., and Shen H.B., 2008, Cell-PLoc: a package of Web servers for predicting subcellular localization of proteins in various organisms, Nat. protoc., 3(2): 153-162
http://dx.doi.org/10.1038/nprot.2007.494 
 
de Castro E., Sigrist C.J., Gattiker A., Bulliard V., Langendijk-Genevaux P.S., Gasteiger E., Bairoch A., and Hulo N., 2006, ScanProsite: detection of PROSITE signature matches and ProRule-associated functional and structural residues in proteins, Nucleic Acids Res., 34(Web Server issue): W362-365

De-la-Peña C., Badri D.V., Lei Z., Watson B.S., Brandão M.M., Silva-Filho M.C., Sumner L.W., and Vivanco J.M., 2010, Root secretion of defense-related proteins is development-dependent and correlated with flowering time, J. Biol. Chem., 285: 30654-30665
http://dx.doi.org/10.1074/jbc.M110.119040 
 
Ding Y., Wang J., Wang J., Stierhof Y.D., Robinson D.G., and Jiang L., 2012, Unconventional protein secretion, Trends Plant Sci., 7: 606-615
http://dx.doi.org/10.1016/j.tplants.2012.06.004 
 
Duvick J., Fu A., Muppirala U., Sabharwal M., Wilkerson M.D., Lawrence C.J., Lushbough C., and Brendel V., 2008, PlantGDB: a resource for comparative plant genomics, Nucl. Acids Res., 36: D959-965
http://dx.doi.org/10.1093/nar/gkm1041 
 
Emanuelsson O., Brunak S., von Heijne G., and Nielsen H., 2007, Locating proteins in the cell using TargetP, SignalP and related tools, Nat. Protoc., 2: 953-971
http://dx.doi.org/10.1038/nprot.2007.131 
 
Emanuelsson O., Nielsen H., Brunak S., and von Heijne G., 2000, Predicting subcellular localization of proteins based on their N-terminal amino acid sequence, J. Mol. Biol., 300: 1005-1016
http://dx.doi.org/10.1006/jmbi.2000.3903 
 
Finnie C., Andersen B., Shahpiri A., and Svensson B., 2011, Proteomes of the barley aleurone layer: A model system for plant signalling and protein secretion, Proteomics, 11: 1595-1605
http://dx.doi.org/10.1002/pmic.201000656 
 
Foresti O., and Denecke J., 2008, Intermediate organelles of the plant secretory pathway: identity and function, Traffic, 9: 1599-1612
http://dx.doi.org/10.1111/j.1600-0854.2008.00791.x 
 
Gillmor C.S., Lukowitz W., Brininstool G., Sedbrook J.C., Hamann T., Poindexter P., and Somerville C., 2005, Glycosylphosphatidylinositol-anchored proteins are required for cell wall synthesis and morphogenesis in Arabidopsis, Plant Cell, 17:1128-1140
http://dx.doi.org/10.1105/tpc.105.031815 
 
Greenbaum D., Luscombe N.M., Jansen R., Qian J., and Gerstein M., 2001, Interrelating different types of genomic data, from proteome to secretome: coming in on function, Genome Res., 11: 1463-1468
http://dx.doi.org/10.1101/gr.207401 
 
Hathout Y., 2007, Approaches to the study of the cell secretome, Expert Rev. Proteomics, 4: 239-248
http://dx.doi.org/10.1586/14789450.4.2.239 
 
Horton P., Park K.J., Obayashi T., Fujita N., Harada H., Adams-Collier C.J., and Nakai K., 2007, WoLF PSORT: protein localization predictor. Nucleic acids res., 35(Web Server issue): W585-587

Isaacson T., and Rose J.K.C., 2006, The plant cell wall proteome, or secretome, In Plant Proteomics, Annual Plant Reviews Series, edited by Finnie C., Blackwell Publishing, 28:185-209

Jamet E., Albenne C., Boudart G., Irshad M., Canut H., and Pont-Lezica R, 2008, Recent advances in plant cell wall proteomics, Proteomics, 8: 893-908
http://dx.doi.org/10.1002/pmic.200700938 
 
Jones R.L., and Robinson D.G., 1989, Protein Secretion in Plants, Tansley Review No. 17, New Phytologist, 111: 567-597
http://dx.doi.org/10.1111/j.1469-8137.1989.tb02352.x 
 
Jung Y.H., Jeong S.H., Kim S.H., Singh R., Lee J.E., Cho Y.S., Agrawal G.K., Rakwal R., and Jwa N.S., 2008, Systematic secretome analyses of rice leaf and seed callus suspension-cultured cells: workflow development and establishment of high-density two-dimensional gel reference maps, J. Proteome Res., 7: 5187-5210
http://dx.doi.org/10.1021/pr8005149 
 
Käll L., Krogh A., and Sonnhammer E.L.L., 2007, Advantages of combined transmembrane topology and signal peptide prediction--the Phobius web server, Nucleic acids res., 35(Web Server issue): W429-432

Kamoun S., 2009, The Secretome of Plant-Associated Fungi and Oomycetes, In: Deising V.H. (ed.), Plant Relationships, 2nd Edition, The Mycota, Springer-Verlag, Berlin Heidelberg, pp 173-180

Kim S.T., Kang Y.H., Wang Y., Wu J., Park Z.Y., Rakwal R., Agrawal G.K., Lee S.Y., and Kang K.Y., 2009, Secretome analysis of differentially induced proteins in rice suspension-cultured cells triggered by rice blast fungus and elicitor, Proteomics, 9: 1302-1313
http://dx.doi.org/10.1002/pmic.200800589 
 
Krause C., Richter S., Knöll C., and Jürgens G., 2013, Plant secretome - From cellular process to biological activity, Biochim. Biophys. Acta, 1834(11): 2429-2441

Krogh A., Larsson B., von Heijne G., and Sonnhammer E.L.L., 2001, Predicting transmembrane protein topology with a hidden Markov model: Application to complete genomes, J. Mol. Biol., 305: 567-580
http://dx.doi.org/10.1006/jmbi.2000.4315 
 
Kusumawati L., Imin N., and Djordjevic M.A., 2008, Characterization of the secretome of suspension cultures of Medicago species reveals proteins important for defense and development, J. Proteome Res., 7: 4508-4520
http://dx.doi.org/10.1021/pr800291z 
 
Lopez-Casado G., Urbanowicz B.R., Damasceno C.M.B., and Rose J.K.C., 2008, Plant glycosyl hydrolases and biofuels: a natural marriage, Current Opinion Plant Biol., 11: 329-337
http://dx.doi.org/10.1016/j.pbi.2008.02.010 
 
Lum G., Vanburen R., Ming R., Min X.J., 2013, Secretome prediction and analysis in sacred lotus (Nelumbo nucifera Gaertn.), Tropical Plant Biol., 6:131-137
http://dx.doi.org/10.1007/s12042-013-9121-5 
 
Lum G., and Min X.J., 2013, Bioinformatic protocols and the knowledge-base for secretomes in fungi, In: Gupta V.K., Tuohy M.G., Ayyachamy M., Turner K.M. and O’Donovan A. (eds.), Laboratory Protocols in Fungal Biology: Current Methods in Fungal Biology, Springer, pp 545-557
http://dx.doi.org/10.1007/978-1-4614-2356-0_54 
 
Lum G., and Min X.J., 2011a, Plant secretomes: Current status and future perspectives, Plant Omics J., 4: 114-119

Lum G., and Min X.J., 2011b, FunSecKB: the fungal secretome knowledgebase, Database - J. Biol. Databases Curation, Vol. 2011
http://dx.doi.org/10.1093/database/bar001 
 
Marchler-Bauer A., Lu S., Anderson J.B., Chitsaz F., Derbyshire M.K., DeWeese-Scott C., Fong J.H., Geer L.Y., Geer R.C., Gonzales N.R., Gwadz M., Hurwitz D.I., Jackson J.D., Ke Z., Lanczycki C.J., Lu F., Marchler G.H., Mullokandov M., Omelchenko M.V., Robertson C.L., Song J.S., Thanki N., Yamashita R.A., Zhang D., Zhang N., Zheng C., and Bryant S.H., 2011, CDD: a Conserved Domain Database for the functional annotation of proteins, Nucleic Acids Res., 39(Database issue): D225-229
http://dx.doi.org/10.1093/nar/gkq1189 
 
McCarthy F.M., Wang N., Magee G.B., Nanduri B., Lawrence M.L., Camon E.B., Barrell D.G.,Hill D.P., Dolan M.E., Williams W.P., Luthe D.S., Bridges S.M., and Burgess S.C., 2006, AgBase: a functional genomics resource for agriculture, BMC Genomics, 7: 229
http://dx.doi.org/10.1186/1471-2164-7-229 
 
Meinken J., and Min X.J., 2012, Computational prediction of protein subcellular locations in eukaryotes: an experience report. Comput. Mole. Biol., 2(1): 1-7

Melhem H., Min X.J., and Butler G., 2013, The impact of SignalP 4.0 on the prediction of secreted proteins, 2013 IEEE Symposium Series on Computational Intelligence (IIEEE SSCI 2013): The 10th annual IEEE Symposium on Computational Intelligence in Bioinformatics and Computational Biology, Singapore, pp.16-22

Min X.J., 2010, Evaluation of computational methods for secreted protein prediction in different eukaryotes, J. Proteomics Bioinform., 3: 143-147

Min X.J., Butler G., Storms R., and Tsang A., 2005a, OrfPredictor: predicting protein-coding regions in EST-derived sequences, Nucleic Acids Res., 33: W677-680
http://dx.doi.org/10.1093/nar/gki394 
 
Min X.J., Butler G., Storms R., and Tsang A., 2005b, TargetIdentifier: a web server for identifying full-length cDNAs from EST sequences, Nucleic Acids Res., 33: W669-672
http://dx.doi.org/10.1093/nar/gki436 
 
Ming R., Vanburen R., Liu Y., Yang M., Han Y., Li L.T., Zhang Q., Kim M.J., Schatz M.C.,Campbell M., Li J., Bowers J.E., Tang H., Lyons E., Ferguson A.A., Narzisi G., Nelson D.R., Blaby-Haas C.E., Gschwend A.R., Jiao Y., Der J.P., Zeng F., Han J., Min X.J., Hudson K.A.,Singh R., Grennan A.K., Karpowicz S.J., Watling J.R., Ito K., Robinson S.A., Hudson M.E., Yu Q., Mockler T.C., Carroll A., Zheng Y., Sunkar R., Jia R., Chen N., Arro J., Wai C.M., Wafula E., Spence A., Han Y., Xu L., Zhang J., Peery R., Haus M.J., Xiong W., Walsh J.A., Wu J., Wang M.L., Zhu Y.J., Paull R.E., Britt A.B., Du C., Downie S.R., Schuler M.A., Michael T.P., Long S.P., Ort D.R., Schopf J.W., Gang D.R., Jiang N., Yandell M., Depamphilis C.W., Merchant S.S., Paterson A.H., Buchanan B.B., Li S., Shen-Miller J., 2013, Genome of the long-living sacred lotus (Nelumbo nucifera Gaertn.), Genome Biol., 14(5): R41
http://dx.doi.org/10.1186/gb-2013-14-5-r41 
 
Ngara R., and Ndimba B.K., 2011, Mapping and characterization of the sorghum cell suspension culture secretome, African J. Biotechnol., 10: 253-266

Oh I.S., Park A.R., Bae M.S., Kwon S.J., Kim Y.S., Lee J.E., Kang N.Y., Lee S., Cheong H., and Park O.K., 2005, Secretome analysis reveals an Arabidopsis lipase involved in defense against Alternaria brassicicola, Plant Cell, 17: 2832-2847
http://dx.doi.org/10.1105/tpc.105.034819 
 
Petersen T.N., Brunak S., von Heijne G., and Nielsen H., 2011, SignalP 4.0: discriminating signal peptides from transmembrane regions, Nature Methods, 8: 785-786
http://dx.doi.org/10.1038/nmeth.1701 
 
Poisson G., Chauve C., Chen X., and Bergeron A., 2007, FragAnchor a large scale all Eukaryota predictor of Glycosylphosphatidylinositol-anchor in protein sequences by qualitative scoring, Genomics Proteomics Bioinform., 5: 121-130
http://dx.doi.org/10.1016/S1672-0229(07)60022-9 
 
Ranki H., and Sopanen T., 1984, Secretion of alpha-amylase by the aleurone layer and the scutellum of germinating barley grain, Plant Physiol., 75: 710-715
http://dx.doi.org/10.1104/pp.75.3.710 
 
Rose J.K., and Lee S.J., 2010, Straying off the highway: trafficking of secreted plant proteins and complexity in the plant cell wall proteome, Plant Physiol., 153: 433-436
http://dx.doi.org/10.1104/pp.110.154872 
 
Shinano T., Komatsu S., Yoshimura T., Tokutake S., Kong F.J., Watanabe T., Wasaki J., Osaki M., 2011, Proteomic analysis of secreted proteins from aseptically grown rice, Phytochemistry, 72: 312-320
http://dx.doi.org/10.1016/j.phytochem.2010.12.006 
 
Sigrist, C.J.A., Cerutti, L., de Casro, E., Langendijk-Genevaux, P.S., Bulliard, V., Bairoch, A., and Hulo N., 2010, PROSITE, a protein domain database for functional characterization and annotation, Nucleic Acids Res., 38: 161-166
http://dx.doi.org/10.1093/nar/gkp885 
 
Simpson C., Thomas C., Findlay K., Bayer E., and Maule A.J., 2009, An Arabidopsis GPI-anchor plasmodesmal neck protein with callose binding activity and potential to regulate cell-to-cell trafficking, Plant Cell, 21: 581-594
http://dx.doi.org/10.1105/tpc.108.060145 
 
Sottomayor M., and Barceló A.R., 2004, Plant peroxidases and phytochemistry – foreword, Phytochemistry Rev., 3: 1-2
http://dx.doi.org/10.1023/B:PHYT.0000047819.21396.15 
 
Tjalsma H., Bolhuis A., Jongbloed J.D., Bron S., and van Dijl J.M., 2000, Signal peptide-dependent protein transport in Bacillus subtilis: a genome-based survey of the secretome, Microbiol. Mol. Biol. Rev., 64: 515-547
http://dx.doi.org/10.1128/MMBR.64.3.515-547.2000 
 
von Heijne G., 1990, The signal peptide, J. Membr. Biol., 115: 195-201
http://dx.doi.org/10.1007/BF01868635 
 
Wen F., VanEtten H.D., Tsaprailis G., and Hawes M.C., 2007, Extracellular proteins in pea root tip and border cell exudates, Plant Physiol., 143: 773-783
http://dx.doi.org/10.1104/pp.106.091637 
 
Werck-Reichhart D., and Feyereisen R., 2000, Cytochromes P450: a success story, Genome Biol., 1: REVIEWS3003
Zhang L., Tian L.H., Zhao J.F., Song Y., Zhang C.J., and Guo Y., 2009, Identification of an apoplastic protein involved in the initial phase of salt stress response in rice root by two-dimensional electrophoresis, Plant Physiol., 149: 916-928
http://dx.doi.org/10.1104/pp.108.131144