Abstract
It is getting familiar that pathway information greatly contributes to elucidate the molecular basis of human disease with large-scale biological data. We developed a pathway database for molecular pathology in periodontitis named 'Pathogenic Pathway Database for Periodontitis'. Periodontitis is an inflammation disease in periodontal tissue and associated with an increased health risk of angina, myocardial infarction, and fetal cardiovascular events. Despite accumulation of biomedical research on periodontitis pathology at the molecular level, there has been no systematization for biological pathways in periodontitis. We checked 185 reference papers and extracted causal relationships among molecules as well as pathways representing molecular etiology. We also built an ontology that systematizes conceptual terms associated with the pathways. Besides pathways, our ontology provides cellular and tissue specific contextual information that is required to represent pathology at the cellular and tissue specific levels. We implemented in our database an integrated viewer for the association between pathways and ontology. Pathogenic Pathway Database for Periodontitis is freely available at http://bio-omix.tmd.ac.jp/disease/perio/.
Keywords: pathway database, ontology, pathogenic pathway, periodontitis, alveolar bone resorption
Introduction
Pathway information is valuable in elucidating the molecular basis of human disease with large-scale biological data such as expression profiling using DNA microarrays. Over Representation Analysis (ORA), such as Gene Set Enrichment Analysis (GSEA) [1], Learner of Functional Enrichment (LEFE) [2], and GeneTrail [3] has been developed for evaluation of microarray data at the level of pathways for association with disease phenotype. ORA identifies pathways that include as many genes as possible with coordinated changes in their expression. This approach has been successful in identifying oxidative phosphorylation as a pathogenic pathway in diabetes II [4] and a regulatory mechanism of mTOR, a promising drug target, in transcription of mitochondrial oxidative function [5].
However, there seem not so many novel discoveries of molecular mechanisms responsible for pathogenesis, while a tremendous amount of DNA microarray data have been measured in a number of laboratories around the world. We consider that there is a certain limitation in the current approach of ORA with pathway knowledge. ORA can detect better scored pathways among given pathways which are usually provided by public databases such as KEGG [6], Reactome [7], and BioCarta [8], or a number of commercial databases like TRANSPATH [9]. Although the public databases cover a wide range of biological pathways, the currently available pathways mostly account for cells not in a pathogenic but in the normal ('healthy') status. In contrast, most of the gene expression data are obtained from cells in a pathogenic status. We start from the idea that a pathogenic cell has the ability to stabilize its pathogenic state by transforming a 'healthy' pathway into a different one that is specific for the pathogenic state [10]. The current public databases contain little information about such pathogenic pathways but mostly contain pathogenic changes (mutations) in individual components of the pathway (genes). We believe that ORA's detection of coordinated alterations in gene expression patterns might more nicely fit with pathways specific to individual disease phenotypes. With the accumulation of disease specific pathways it is expected to obtain better results from ORA analysis.
We are motivated to evaluate the differences between healthy and pathogenic cellular states at the pathway level. Here we report on the development of a pathway database for molecular pathology in periodontitis named 'Pathogenic Pathway Database for Periodontitis'. We have systematized information about pathways that account for periodontitis characterized by alveolar bone resorption. Periodontitis is an inflammation disease that affects the periodontal tissue and is associated with an increased health risk of angina, myocardial infarction, and fetal cardiovascular events. Clinical characteristics of this disease include periodontal pocket formation and alveolar bone resorption in conjunction with the inflammation. Periodontitis is known to be a multifactor disease caused by immune factors, microbiological factors, environmental factors, genetic factors, systemic disease, drug and chemical factors and occlusion. We collected flowcharts that illustrate pathogenic pathways starting at various multifactors and ending up with alveolar bone disorder in periodontitis.
As far as we know this is the first example of development of a pathway database for periodontitis; however, existing pathway databases already contain certain amount of pathway information on pathogenic states. Reactome [7] collects pathway information on bacterial or viral infections such as Botulinum neurotoxicity, HIV infection, and influenza infection. KEGG [6] contributes pathway information on four major prevailing diseases: neurodegenerative disorders, cancer, infectious diseases, and metabolic disorders. The Comparative Toxicogenomics Database (CTD) [11] is a recently emerging database that provides information on molecular mechanisms by which environmental chemicals affect human diseases. CTD provides periodontitis related information: 18 chemicals affecting progression of periodontitis as well as 25 interactions between these chemicals and proteins. Unfortunately CTD currently provides no pathway information on periodontitis.
In our development of a pathway database for molecular pathology in periodontitis, we consider that we need to include information not only on molecular reactions but also on cellular and anatomical environments where molecular reactions occur. We applied 1) an ontology and 2) a semi-structured description in XML to integrate the contextual information with pathways. We developed an ontology for the molecular pathology of periodontitis in order to systematize multiple factors other than molecules that contribute to pathogenesis of periodontitis. On the basis of our ontology we marked up conceptual terms in every causal relationship between biological entities by XML tags. This paper shows the ontology and the XML description work in representing cellular and tissue specific information accounting for the pathogenic status in periodontitis.
Methods and results
Data collection
We manually extracted biological data on the molecular etiology of periodontitis from PubMed abstracts. We investigated 185 abstracts collected by keyword search for 'periodontitis and osteoclast' in PubMed on July 31, 2008. The extracted biological data consist of causal relationships between biological entities, species, cell types, and experimental methods. We collected manually 698 causal relationships from 185 abstracts.
XML semi-structured description of causal relationships data
We represented causal relationships information in plain text. In order to clarify semantics of the information, we marked up conceptual terms in the text by XML tags. We defined 16 XML tags (Tab. 1) for marking up the terms. The XML tags consist of markers for causal molecules as well as cellular and anatomical environments where the molecules act (Tab. 2). The entire data of causal relationships are available in XML as Supplementary Data.
| Table 1: XML tags and optional attributes used in semi-structured representations of causal relationships data in our database. |
| XML tag | specification |
| relation | a causal relationship |
| pathID | an internal ID of a causal relationship |
| refID | an internal ID of a reference paper |
| pubmed | a PubMed ID of a reference paper |
| disease | a disease |
| from | a causal part of a binary relationship |
| to | a consequent part of a binary relationship |
| bio | a molecule (e.g. gene, protein, metabolite) originating in a human cell |
| bacteria | a molecule (e.g. gene, protein, metabolite) originating in bacteria |
| chemical | a chemical and drug |
| cell | a cell |
| organ | an organ |
| organism | a species |
| complex | a complex of biological objects (e.g. intracellular signaling cascade) |
| phenomena | a phenomenon including both a 'biological phenomenon' and a 'pathological condition, sign and symptom' |
| property | a qualifier to the other marked up conceptual terms |
| Table 2: | An example for descriptions of causal relationships in XML*. |
| <relation> |
| <pathID>138</pathID> |
| <refID>19</refID> |
| <pubmed>17041006</pubmed> |
| <disease>periodontitis</disease> |
| <from><organism>Gram-negative bacteria</organism> derived <bacteria>Lipopolysaccharide</bacteria></from> |
| <to>initiate <property>inflammatory</property> <phenomena>bone loss</phenomena></to> |
| </relation> |
| * Individual XML-tags are specified in Tab. 1. In this entry (ID=138), 'from' element includes two marked-up conceptual terms ('Gram-negative bacteria' and 'Lipopolysaccharide') as well as description in plane text, while 'to' element includes tew marked-up conceptual term ('inflammatory' and 'bone loss') as well as description in plane text. |
Ontology for molecular pathology of periodontitis
Both XML tags in Tab. 1 and all the marked-up conceptual terms were taken up into our ontology. In defining its higher concepts, we refer to Mizoguchi's top ontology [12] and its application to biological knowledge [13, 14]. We believe this is the first ontology systematizing the molecular pathology of periodontitis. The ontology consists of 404 conceptual terms which are largely grouped into two top categories of physical thing and abstract thing. The top category physical thing subsumes continuant and occurrent (Fig. 1).
 Click on the thumbnail to enlarge the picture |
Figure 1: A hierarchy of top categories in our ontology for molecular pathology in periodontitis. |
The category continuant subsumes molecule, cell, organ, living organism, complex of molecules, and complex of living organisms. The class molecule represents molecular factors including genes, proteins, metabolites and xenobiotics. The class cell represents cells where various cytokines or chemokines are produced and secreted. The class organ represents organs of the human body. Many of periodontium related organs belong to the organ class in our ontology. We defined subclasses of cell and organ according to the conceptual hierarchy of MeSH ontology in NCBI. The class living organism represents living things such as animals, plants, insects, and bacteria, although this ontology includes only human and bacteria as subclasses of living organism. The class complex of molecules represents physical objects defined by certain combinations of different kinds of molecules (e.g. 'intracellular signaling cascade' and 'inflammatory infiltrate'). Similarly the class complex of living organisms represents physical objects defined by combinations of different kinds of living organisms (e.g. 'biofilm microorganisms').
Periodontitis distinguishes itself by involving various microorganisms in its molecular etiology. Many molecules derived from microorganisms play important roles in the pathogenesis. Lipopolysaccharide (LPS) and lipooligosaccharide (LOS) are some examples. Periodontitis can also be characterized by the specific drugs used in its treatment. Because of these characteristics, our ontology contains three subcategories under molecule: 1) the class biological molecule, which includes molecules originating in human cells (e.g. 'interleukin'), 2) the class bacterial molecule, which includes molecules originating in bacteria and acting as pathogens (e.g. 'lipopolysaccharide'), and 3) the class chemical and drug, which includes chemical compounds acting as therapeutic drugs or pathogenic factors in concert with bacteria (e.g. 'bisphosphonate' and 'nicotine').
Besides continuant, occurrent subsumes two subclasses according to Mizoguchi's top ontology [12]: state and phenomenon. We classified disease as a subclass of state. Phenomenon subsumes two subcategories: 1) the class biological phenomenon, which represents phenomena observed in healthy cells and organs (e.g. 'angiogenesis' and 'bone resorption') and 2) the class pathogenic condition, sign and symptom, which represents the phenomena characterizing pathogenic states of cells and organs (e.g. 'bacterial insult' and 'resorption pit') (Fig. 1).
The other top category abstract thing subsumes quality, and then property in sequence as its subcategories. The class property represents conceptual terms that qualify other conceptual terms belong to physical thing category. For example, inflammatory belonging to the property class qualifies bone loss belonging to the phenomenon class (Tab. 2).
Pathway graphs
We constructed 20 pathways by grouping biological relationships data. We then manually drew 20 pathway graphs that show flowcharts of the molecular etiology for alveolar bone resorption in periodontitis. The 20 individual pathways illustrate 1) Aggregatibacter actinomycetemcomitans (A. a.) infection, 2) Porphyromonas gingivalis (P. g.) infection, 3) Treponema denticola (T. d.) infection, 4) adiponectine influence, 5) baicalin treatment, 6) bisphosphonate treatment, 7) cemetidine treatment, 8) conjugated linoleic acid (CLA) treatment, 9) capsular-like polysaccharide antigen (CPZ) treatment, 10) DHA treatment, 11) indomethacin treatment, 12) kariotoxin influence, 13) parthenolide (PAR) treatment, 14) parathyroid hormone (PHT) treatment, 15) polymyxin B treatment, 16) taurine treatment, 17) tetracyclines treatment, 18) thalidomide treatment, 19) smoking (nicotine) influence, and 20) type II diabetes influence.
Our pathway graphs consist of four kinds of edges and seven kinds of nodes: edge-1) intercellular relations with active effect (pale blue and broken lines with arrow heads), edge-2) intracellular relations with active effect (pale blue and solid lines with arrow heads), edge-3) any relation with suppressive effect (orange and broken lines with blunt heads), edge-4) components of a cell (pale blue and solid lines without heads), node-1) secreted molecules originating from human cells (pale blue ovals with dotted borders), node-2) non-secreted intracellular molecules originating from human cells (pale blue ovals without borders), node-3) molecules originating from bacteria (pink ovals without borders), node-4) cells (yellowish green ovals without borders), node-5) bacteria (purple ovals without borders), node-6) biological phenomena (yellow rectangles rounded off their corners without borders), and node-7) chemical and drug (blue rectangles rounded off their corners without borders). In addition, numbers attached to edges indicate internal IDs of reference papers in our database. Clicking on the numbers in the graphs invokes details of causal relationship information extracted from the reference papers as well as hypertext links to their PubMed entries (Fig. 2).
We also manually depicted a pathway that gives an overview of the 20 pathways (Fig. 3). The overview categorizes the 20 pathways on the basis of pathological multi-factors of periodontitis: immune factors, microbiological factors, environmental factors, genetic factors, systemic disease, drug and chemical factors and occlusion. Because these factors are familiar to researchers expert in periodontitis, Fig. 3 helps the researchers to understand how their background knowledge relates to the pathways provided in our database.
 Click on the thumbnail to enlarge the picture |
Figure 2: An example of pathway graphs. It shows flowcharts of pathogenic pathways starting at Porphyromonas gingivalis (P. g.) infection and ending up with osteoclastogenesis and bone resorption. |
 Click on the thumbnail to enlarge the picture |
Figure 3: A pathway that gives an overview of all the 20 pathways contained in our database. Seven green nodes indicate major pathogenic factors of periodontitis, which is indicated by blue edges. While pale blue edges indicate biological regulations between the pathogenic factors. Individual 20 pathways are marked associated with their pathogenic factors in this figure. |
Web interface
We designed our database as a Web database. We implemented application programs in Perl to manipulate users' requests against causal relationships data as well as the ontology, both of which are stored in plain text. We do not use a database management system because of the rather small size of the data (968 entries). The application programs provide users the following facilities: 1) key word search for marked-up terms and phrases by XML-tags in causal relationships data (Fig. 4a), 2) a pathway browser that provides hypertext links from a pathway graph to individual causal relationships data describing the details about the pathway (Fig. 4b), 3) an ontology browser that provides a tree-like view for the ontology in which an individual leaf-node has a link to causal relationships data (Fig. 4c), 4) table-style representation of causal relationships data in XML that provides a link from a marked-up gene name to a corresponding entry of EntrezGene in NCBI as well as a link from a PubMed ID to a corresponding entry of PubMed in NCBI (Fig. 4d).
 Click on the thumbnail to enlarge the picture |
Figure 4: Upper panel: Application programs that guide a user to entries in our Pathogenic Pathway Database for Periodontitis. Lower panel: An instance of view in our web interface. This view consists of (c) a tree-like view of our ontology with a highlight at the conceptual term of 'infection', (d) causal relationships data associated with 'infection', (b) a view of 'bisphosphonate pathway' in which a node has a link to a causal relationship data in (d), and (a) keyword query. |
Integration of Pathway Graph and Ontology
Diseases cannot be described only by molecular interactions. Biological phenomena and cellular and tissue specific contexts are required for a complete description of disease states. However, it is very difficult to include in a pathway graph other information than molecular relationships mostly because of limitation in the depicting space. We then included all the contextual information required to illustrate pathogenesis in our ontology and tried to integrate it with the pathway graphs.
For this purpose, we implemented an ontology viewer associated with the pathway graphs. Fig. 5 shows an example illustrating 'T.d. (Treponema denticola) pathway' (Fig. 5a) and a view of that part of the ontology that is associated with this pathway (Fig. 5b). While the entire ontology represents all the biological entities contributing to periodontitis (Fig. 1), a part of the ontology in Fig. 5 accounts for cellular and tissue specific contexts illustrating 'T.d. pathway'. The partial ontology indicates that T.d. pathway occurs in osteoclast cells of gingival tissue upon infection by Treponema denticola or Treponema socranskii bacteria with the consequence of osteoclastogenesis and periodontitis. These conceptual terms indicated in bold letters illustrate contextual information in time and space where molecular interactions in a pathway occur.
 Click on the thumbnail to enlarge the picture |
Figure 5: An example of views provided by our ontology viewer associated with the pathway graph. a: The pathway graph of 'T.d. pathway' that describes a flowchart of pathogenic influence of T.d. bacteria ending up with osteoclastogenesis. b: A part of our ontology that accounts for cellular and tissue specific information illustrating pathogenesis of 'T.d. pathway'. In this figure, a square filled by a plus-letter indicates our omission of names of molecules that follow 'molecule' category from this figure. |
Discussion
Ontology associated with pathway databases
There are pathway databases including ontologies for pathways. KEGG [6] and BioCyc [15] include ontologies that classify pathways contained in the databases. However both these ontologies do not specify factors associated with the pathways such as cellular and tissue specific contextual factors as well as biological and/or pathological phenomena. This type of classifications provided by KEGG and BioCyc plays a similar role as the Biological Process ontology provided by the Gene Ontology consortium [16], which specifies a 'process' itself. In contrast our ontology specifies individual components in a 'process', in order to specify the molecular mechanisms underlying pathogenesis in periodontitis.
There are also ontologies for specification of biological pathways: BioPAX [17], Signal Ontology [13, 14], Event ontology [18], and Cell System Ontology [19]. While these ontologies specify biochemical reactions and their biological interpretations, our ontology specifies cellular and tissue specific information explicating why the biochemical reactions contribute to pathogenesis of a disease.
Pathway information on bone resorption
Periodontitis is a multi-factor disease and many biological processes participate in its progression. Among those we focused on 'alveolar bone resorption'. This process has close relations with a bone resorption process in other diseases caused by inflammation of bone tissue such as rheumatoid arthritis. We found pathway information on bone resorption provided in BioCarta [8] in an entry of 'Bone Remodeling'. We investigated the difference between pathways of BioCarta and our database. We found at first that our pathways include more detailed information on the molecular pathogenesis as well as more biological molecules playing roles in the pathogenesis. Second, our pathway specifies pathological contributions of bacteria as well as pathogenic molecules from the bacteria. In addition, our pathways also specify the influence of various chemical compounds on bone remodeling. Comparative Toxicogenomics Database (CTD) [11] also includes a similar list of chemical compounds used as drugs for alveolar bone resorption; however CTD does neither include information about pathways nor about molecular responses induced by drugs.
Systematization of disease related concepts in ontology
We compared our ontology for periodontitis with the MeSH ontology of NCBI. Alveolar Bone Remodeling in Periodontitis appears in MeSH under the class Periodontal Atrophy, which follows Periodontal Diseases going up to Diseases Category. MeSH ontology includes most of all the conceptual terms in our ontology such as gingival tissue, osteoblast, and Aggregatibacter actinomycetemcomitans (A.a.). We then regard our ontology as a sort of subset of MeSH ontology that specifies the molecular pathology in periodontitis. Taking into account the large size and complexity of MeSH ontology, it is worth to define a subset of the ontology that specifies molecular pathology of individual diseases, as our ontology shows.
Future works
In the next step we will apply our pathway collection to analyze gene expression profiling data from periodontitis patients. There are publicly available microarray data in GEO [20]: GSE13042, GSE12484, GSE6751, GSE2525 GSE10334, GSE13903, and GSE341, on which no pathway-related analysis has been conducted so far, to the best of our knowledge. We will evaluate these microarray data with our pathway information about how strongly our pathways contribute to the disease phenotypes. The analysis will be able to evaluate the difference between bone-resorption pathways specific to periodontal tissue and bone-resorption pathways common to many tissues and organs. We also plan to investigate original DNA microarray data provided by our collaborators.
In addition we will expand our database development for pathways of various pathogenic phenotypes. We will target adult diseases characterized by multiple risk factors such as hypertension, type II diabetes, and congestive heart failure and stroke. Accumulation of pathway information will support rapid development of bioinformatics that can elucidate novel molecular pathology for multifactor diseases.
Acknowledgements
The authors are grateful to Prof. Edgar Wingender (Goettingen University) for his insightful comments on this manuscript.