If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Divisions of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, Jefferson, ArkansasNational Center for Safety Evaluation of Drugs, National Institute for Food and Drug Control, Beijing, China
Drug-induced liver injury (DILI) may present any morphologic characteristic of acute or chronic liver disease with no standardized terminology in place. Defining lexemes of DILI histopathology would allow the development of advanced knowledge discovery and data mining tools for across comparisons of publicly available information. For these purposes, a DILI ontology (DILIo) was developed by using the Unified Medical Language System tool and the standardized terminology of the Systematized Nomenclature of Medicine–Clinical Terms (SNOMED CT). The DILIo was entrained on findings of 114 US Food and Drug Administration–approved drugs by extracting all clinically DILI-related histopathologic descriptions for 1082 liver biopsy samples, which were then analyzed using the Unified Medical Language System MetaMap and subsequently mapped to the SNOMED CT. The DILIo provides a standard means to describe and organize liver injury induced by drugs, enabling comparative analysis of drugs within and across histopathologic terms. The analysis showed that flutamide, troglitazone, diclofenac, isoniazid, and tamoxifen were reported to have the most diverse histopathologic observations in liver biopsy. Necrosis, cholestasis, fatty degeneration, fibrosis, infiltrate, and hepatic necrosis were the most frequent terms used as descriptors of histopathologic features of DILI. In conclusion, DILIo entrains different algorithms for an efficient meta-analysis of published findings for an improved understanding of mechanisms and clinical characteristics of DILI.
Drug-induced liver injury (DILI) is a major challenge for clinical practice, drug development, and regulation. DILI remains a major adverse drug event that causes the termination of drug development programs and results in regulatory actions, such as drug withdrawals and black box label warnings.
Studying drugs that cause the same histopathologic changes and performing comparative analyses of drugs with different histopathologic manifestations could help define specific features of liver injury and guide further research for the prevention of DILI.
However, researchers report histopathologic findings in free text, which has resulted in diverse terms being used to describe the same injury in clinical reports. For example, steatosis and fatty degeneration are both terms used to describe the same type of injury. Such relaxed use of vocabulary makes the interpretation and use of histopathologic data difficult, especially for researchers without specialized training in pathology. Therefore, a standardized approach to categorize histologic changes in the liver is needed to improve our understanding of liver injury and assist in developing effective prevention strategies.
these data may or may not match with morphologic findings. For example, an increase in serum bilirubin concentration may not be recapitulated by obstruction of the biliary tree. Similarly, massive changes in liver function test results, such as a >10-fold increase in serum alanine levels, aspartate aminotransferase activity, or lactate dehydrogenase levels evidencing cytolytic hepatitis, may not be seen histopathologically as determined by destruction of the liver parenchyma and architecture. Discrepancy in the grading of severity of histopathologic injury is not uncommon, particularly for assigning causality.
Because the histopathologic features of DILI are often nonspecific and can mimic other acute and chronic liver disease, it is difficult to identify hallmarks for DILI, especially in complex co-medications.
Moreover, there is no standardized histologic scoring system for DILI, thus rendering an integration of information from liver biopsy results into a coherent set of data almost impossible. A standardized DILI assessment is, therefore, needed.
Biological ontologies have emerged as a means of representing and organizing biological concepts in precisely the way that is needed for the study of DILI. This organization and systematization enables biologists, bioinformaticians, and others to derive meaning from large data sets. Constructing an ontology that encompasses the histopathologic terms related to DILI has the potential to transform the unsystematic, free-text histopathology reports that are generated by studying DILI into ordered and useful information.
An ontology is an explicit specification of some concepts, usually based on controlled vocabulary, and their logical relations, usually hierarchical in nature, for a particular subject area.
The accumulation of large-scale experimental and computational data is frequently noted and has led to concomitant growth in the scale and complexity of biological databases. To take these advantages, ontologies are increasingly being used in health care and life sciences applications.
Several ontologies have been developed to foster meta-analysis and knowledge base development and have been applied for biomedical research community use and in drug discovery and development for pharmaceutical product development and manufacturing.
To overcome unmet needs, this study reports the development of a DILI ontology (DILIo) to help integrate and organize the various histopathologic features of DILI and to facilitate the study of mechanisms underlying those clinical findings by bridging biological process and computational informatics. To the best of our knowledge, this is the first development of an ontology for DILI.
Materials and Methods
We analyzed 114 drugs from the previously established liver toxicity knowledge database (LTKB).
These drugs were either withdrawn from the market due to hepatotoxicity or had indications of DILI in one of three US Food and Drug Administration (FDA)–approved drug-labeling sections (ie, boxed warning, warnings and precautions, and adverse reactions). A detailed summary of the drugs used in this study is provided in Supplemental Table S1: 25 drugs that were either withdrawn from the market or had a boxed warning (ie, black box warning) on the label, 60 drugs with indications of DILI described in the warnings and precautions section of the label, and 29 drugs with indications of DILI in the adverse reactions section of the label. We used these drugs as a representative set of data to develop the DILIo and to evaluate their association with different histopathologic liver injury patterns.
For each of these 114 drugs, a PubMed keyword search was conducted to identify in the literature original clinical case reports on liver biopsy in patients with DILI. At least one public study for each drug was collected. Histopathologic observations generated by trained pathologists were extracted from each publication. The following search query was used: [drug name] AND (hepatotoxicity OR liver injury) AND (histopathology OR liver biopsy).
The SNOMED CT
The Systematized Nomenclature of Medicine–Clinical Terms (SNOMED CT) is a standardized, multilingual vocabulary of clinical terminology composed of a systematically organized computer-processable collection of medical terms. It provides codes, terms, synonyms, and definitions covering diseases, findings, procedures, microorganisms, substances, etc. It allows for a consistent way to index, store, retrieve, and aggregate clinical data across specialties and sites of care. The SNOMED CT is one of a suite of designated standards for use in the US federal government systems for the electronic exchange of clinical health information and used by physicians and other health care providers for the electronic exchange of clinical health information. It is the most comprehensive multilingual clinical health care terminology in the world and contributes to the improvement of patient care by underpinning the development of electronic health records that record clinical information in ways that enable meaning-based retrieval. The SNOMED CT provides effective access to information required for decision support and consistent reporting and analysis (http://www.nlm.nih.gov/research/umls/Snomed/snomed_main.html, last accessed June 5, 2012).
was created based on three data sources: i) drugs identified as causes of DILI in three major DILI registries; ii) drugs identified as causes of drug-induced acute liver failure (ALF) in six different data sources, including major ALF registries and previously published ALF studies; and iii) drugs identified as being subjected to serious governmental regulatory actions in Europe or the United States owing to their hepatotoxicity. The Suzuki et al database
consists of 319 drugs determined to be causative agents of hepatotoxicity in adjudicated or well-vetted cases obtained from multiple studies and drugs resulting in serious regulatory actions owing to their hepatotoxicity. Of the 319 drugs, 97 were associated with a disproportionately higher reporting frequency of overall liver injury, and the rest were associated with at least one case of ALF.
used a knowledge-based expert system designed to assess the potential toxicity of a chemical from its structure to identify 626 compounds which were assigned to four categories of hepatotoxicity: evidence of human hepatotoxicity, weak evidence (<10 case reports) of human hepatotoxicity, evidence of animal hepatotoxicity (not tested in humans), and no evidence of hepatotoxicity in any species.
was developed by the FDA’s National Center for Toxicological Research as a part of the LTKB project. Currently, the LTKB contains 287 drugs with a DILI annotation based on FDA-approved drug labels. The drugs are classified into three categories: most-DILI-concern, less-DILI-concern, and no-DILI-concern.
The most-DILI-concern group contains drugs that are withdrawn from the market owing to hepatotoxicity, are given a boxed warning (ie, black box warning) for their potential to cause liver injury, or have a warnings and precautions section that specifies concern about DILI with greater-than-moderate severity. The less-DILI-concern group contains drugs that specify a DILI concern in the warnings and precautions section with low severity or in the adverse reactions section. The no-DILI-concern group is composed of drugs with no DILI description mentioned on the label.
Development of the DILIo
The overall workflow of DILIo development is illustrated in Figure 1. Briefly, it comprises two major processes. For each drug, papers were collected using the query described in Data Collection. Then, the sentences describing the histopathologic findings were extracted by reading. Finally, the histopathologic terms were mapped onto the SNOMED CT by using the Unified Medical Language System MetaMap tools. For all the drugs, the mapped SNOMED CT terms were combined to develop a SNOMED tree by extracting from the SNOMED ontology. Then, the trimming process was undertaken by using deep search theory with short pathway criterion and manual correction. As a result, the DILIos consisted of two main branches: body structure and clinical finding.
The DILIo is entrained on 281 published histopathology reports with findings mapped onto the SNOMED CT using a Unified Medical Language System and the MetaMap tool. Subsequently, a SNOMED ontology tree was constructed as summarized in Figure 1. Histopathology reports of liver biopsies were considered only in cases in which DILI was caused by a single drug (monotherapy) or in which causality assessment identified the culprit drug with certainty. Sex and age information was identified in 579 patients who underwent biopsy [214 male patients (36.96%) and 365 female patients (63.04%)]. The average age was 50.8 and 46.9 years for males and females, respectively. Figure 2 informs the number of DILI cases for individual drugs, with thioguanine representing >140 histopathology reports of liver biopsies. The top five drugs with >50 case entries are thioguanine, amoxicillin, methotrexate, isoniazid, and troglitazone. Histopathologic observations as described in the original articles were extracted and analyzed by MetaMap and subsequently searched against the SNOMED CT vocabulary. This approach collapsed different descriptions of the same histopathologic manifestation into 175 standard terms by the SNOMED. Based on the SNOMED hierarchy, the DILIo was constructed to represent each descriptor with a unique term and to lay out the relationships between terms of differing levels of specificity, such as fibrosis and diffuse fibrosis (Supplemental Figure S1).
The structure of the DILIo can be described in terms of a graph, where each term is a node and the relationships between the terms are edges. Every term in the DILIo has a standardized name (eg, cellular necrosis, inflammation, and cholestasis) and a unique zero-padded identifier prefixed by DILIo (eg, DILIo: 0027540 and DILIo: 0032482). The numerical portion of the ID has no inherent meaning or relation to the position of the term in the ontologies; each characterizes a specific type of liver injury. Two linked terms in the DILIo are in an is-a relationship, which means that one term is a subtype of the other term. The ontology is a hierarchical structure with child terms being more specialized and parents terms less specialized; the lower a term is in the DILIo hierarchy, the more specific the term. The relationships encompassed in the DILIo are directed. For example, focal necrosis is a necrosis, but necrosis is not necessary a focal necrosis. All terms in a domain can trace their parentage to the root term, although there may be multiple paths via varying numbers of intermediary terms to the ontology root.
Among standard terms, the SNOMED uses the vocabularies body structure and clinical finding (Supplmental Figure S1). They represent DILI in two different aspects. The former descriptor is associated with the morphologic findings, and the latter is related to the clinical features. Accordingly, the DILIo contains two major branches: body structure and clinical finding. Both branches have approximately 10 hierarchical levels, with multiple terms at each level. For improved understanding of the association between drugs and morphologic and clinical findings, two types of coding (ie, color and size) were used, where color is an indication of the number of drugs falling in these terms (empty for 0 drugs, green for 1 to 10 drugs, yellow for 11 to 20 drugs, blue for 21 to 30 drugs, and red for >30 drugs) and size provides further resolution, eg, larger in size means more drugs falling in this category specified by the color.
For the body structure branch, 76 morphologic abnormality terms were identified. The branch displayed these 76 structural abnormalities in terms of their logical relationships and the controlled vocabulary in the SNOMED CT. It formed a hierarchical structure (ie, ontology tree) topped by apoptosis and morphologic alterations. Necrosis, fatty degeneration, fibrosis, and infiltrate are dominant terms. Furthermore, bridging and focal and massive necrosis are the most frequently observed necrotic observations in DILI.
Compared with the structural branch, the clinical finding branch is more complex, consisting of more nodes and forming a more complicated network, as depicted in Supplemental Figure S1. More than 50 of 114 drugs are reported for cholestasis, a DILI type, classified according to clinical biochemistry. Hepatic necrosis is another major term after cholestasis. Acute hepatic necrosis is the most common hepatic necrosis, followed by subacute and focal hepatic necrosis. It seems that the number of drugs per term is more condensed in the body structure branch than in the clinical finding branch.
There are five drugs with >10 terms used to describe pathologic findings (Figure 3). These drugs, flutamide, troglitazone, diclofenac, isoniazid, and tamoxifen, are well-known causative agents reported to induce severe liver injury in clinical practice. The clinical characteristics of liver injury based on three databases (Suzuki et al,
) for those five drugs are listed in Table 1. This finding demonstrated that the pathogenesis of DILI usually involves the participation of multiple pathways and indicated that severity of DILI could be associated with the diversity of histopathologic alterations.
Table 1Characteristics of Five Drugs with >10 Terms Reported in the Description of Histopathologic Features
Figure 4 depicts the two-way cluster analysis and shows that necrosis, cholestasis, fatty degeneration, fibrosis, infiltrate, and hepatic necrosis are the most frequent terms used as descriptors of histopathologic features. These six common terms are color-coded in red (necrosis, fatty degeneration, fibrosis, and cholestasis) and blue (infiltrate and hepatic necrosis), as shown in Supplemental Figure S1. There are 29 drugs reported in a combination of three of six dominant histopathologic terms. The clinical presentations of these 29 drugs are summarized in Table 2 and are based on causality assessment summarized in the Suzuki et al,
databases. By identifying the correlation with clinical characteristics of DILI, we found that each of the 29 drugs is well defined in at least one of three databases to induce irreversible, even fatal, liver injury or lead to liver transplantation. The only exception is nevirapine, a first-line antiretroviral regimen that was not found in any of the databases. Its ability to cause hepatotoxicity is of primary concern,
DILI is of great concern to the pharmaceutical industry and regulatory authority. Owing to the plethora of histopathologic manifestations and different mechanisms underlying hepatotoxicity, diverse terms have been used to describe similar morphologic patterns of injuries and the associated clinical manifestations. As of today, a comprehensive ontology has not been developed to evaluate the histologic patterns of DILI. This prompted our interest in evaluating 281 published articles that informed on 1082 original clinical DILI case reports substantiated by liver biopsies. It is important to note that there are several available databases containing information about DILI from diverse perspectives (predictive models, spontaneous reporting systems, the DILI Network, and ALF groups).
However, there is no single database containing such comprehensive histopathologic information as developed in this study to date.
Owing to its complexity and broad spectrum in histopathologic presentations, the use of formal classification to define the DILIo offers an opportunity to integrate knowledge and research findings that have previously not shared any common terms. We considered the need to include terms not only on body structure alterations but also on clinical findings to maximize the coverage of histopathologic observations and enhance the application for future research. Thus, the DILIo consists of two major branches: body structure and clinical finding, as defined by the SNOMED CT. The body structure branch seemed more condensed, with fewer nodes than the clinical finding branch, indicating more diversity of terminology in clinical practice.
Because liver biopsies are not performed routinely, little information is available on the correlation between histopathologic features and clinical manifestations that might inform on mechanisms underlying DILI. With the comprehensive data collection and systematic data analysis of DILIo, we demonstrate that the various histopathologic features present in DILI are an indication of the pathogenesis of DILI and usually involve multiple pathways.
Such pathology confirmed that DILI can be apart from its clinical manifestation, accounted for underlying disease, and provided a clue to mechanisms of injury.
DILIo defined major terms to unify histopathology in DILI across different studies. Necrosis, cholestasis, fatty degeneration, fibrosis, infiltrate, and hepatic necrosis are six major features that dominated the reported observations in liver biopsy of DILI cases. Furthermore, DILIo defined 29 drugs well known for their characteristics of liver injury as causative agents of severe liver injury. Among those drugs, 27 (93%) were also reported as human hepatotoxicants in the Greene et al
Nevirapine is the only exception that is not present in any of the three databases. Being a first-line antiretroviral regimen, its use has been associated with severe hepatotoxicity as reported in various clinical studies and in some cases was associated with an early (<12 weeks) hypersensitivity reaction with a 10.9% incidence of severe hepatotoxicity.
Through the development and systematic analysis of DILIo, we generated the consensus that the severity of DILI is related to the diversity and nature of histopathologic alterations. The present observations agree well with current thinking of the multihit hypothesis in DILI that has been proposed by several investigators.
Liver injury described with >10 terms or presented as the combination of three of six pathology findings adds further weight to the multihit hypothesis, although a single descriptor, such as necrosis, may also be associated with severe DILI if a significant portion of the liver is affected. To the best of our knowledge, this is the first approach to apply the systematic nomenclature of medicine-clinical terms using lexemes for DILI. We hope that this systemization will help the research community, clinical practice, and regulatory science to better analyze published reports, thereby protecting public health from the impact of DILI. Through the development of biomarkers that can be correlated with histologic features, DILI prevention may become feasible in clinical practice.
Exploring the association between drugs and histopathologic presentations can help identify commonality of drugs that cause specific injuries. For example, by grouping drugs according to their shared histopathologic characteristics in the DILIo, one can conduct a meta-analysis of the literature or apply emerging molecular technologies using the data description in the DILIo to evaluate the relationship of the mechanism, drug class, chemical structure, and metabolism with specific histopathologic terms. Results generated from such studies will play a critical role in understanding underlying mechanisms of DILI and, thus, in predicting DILI prognosis and providing guidelines in clinical practice. For example, advances in genetic research and molecular biology techniques have allowed researchers to begin to characterize the genetic components underlying some serious susceptibilities to clinically significant hepatotoxicity that are being increasingly identified.
The identification and validation of these genetic markers will require a large and sufficiently diverse patient database and standardized means of defining phenotypes and genotypes. To meet this challenge, researchers around the world have begun to assemble cohorts of patients who have experienced DILI.
Ongoing efforts to link comprehensive electronic medical records with archived biological material may accelerate the identification and recruitment of these patients. To facilitate cross-study comparisons and collaborations, it is important that the phenotypes of DILI be standardized in their definitions, the optimal phenotype data, and the process for causality assessment. Along with updating and enriching the annotation of current drugs, the DILIo can become a powerful tool in collecting novel research data and enhancing broader application with its standardized, vocabulary-controlled database.
Histopathologic analysis plays an important role in identifying the injury type and the degree of injury in DILI.
The development of a histopathologic DILIo provides standard terms to describe the clinical manifestations of DILI along with a hierarchical relationship between these terms. Thus, the ontology is useful as a tool to survey diverse data and knowledge published in the literature via meta-analysis to draw conclusions that are otherwise not apparent in individual studies. It is hoped that the proposed DILIo will facilitate an automated assessment of published findings and the search for correlations of the histopathologic characteristics with the corresponding clinical outcome. Finally, the ontology can also help design high-throughput experiments to link molecular-level data with hierarchical histopathology findings for biomarker discovery.
The DILIo development provides a representation of biological and clinical knowledge through the use of precisely defined, interrelated terms to support and enhance the analysis of high-throughput data and allow annotation for biological events relevant for the development of mathematical models and computer simulations. By matching with a defined pattern and through a semantic analysis and characterization of morphologic changes of DILI, the diagnosis and confirmation of liver liabilities in complex cases becomes feasible and may even suggest a mechanism(s) of DILI. The severity of DILI is related to the diversity and nature of histopathologic alterations. For DILIo to successfully underpin knowledge-driven database searches and research, a key challenge for long-term maintenance of the DILIo consists of informing it with new knowledge associated with DILI to enrich its application.
Hierarchical trees of the DILIo. Body structure: necrosis, fatty degeneration, and fibrosis are presented in black circles and indicate the most histopathologic alterations in DILI. Necrosis types are focal, bridging, confluent, and massive. Clinical finding: cholestasis is the most observed pathologic feature of DILI, followed by hepatic necrosis and inflammation.
Z. Lin works at the National Center for Toxicological Research through the International Scientist Exchange Program sponsored by the FDA’s Office of International Programs . The LTKB project is supported by the FDA’s Critical Path Initiative , the Office of Women’s Health , and a Chief Scientist Challenge grant. J.B. is supported by the German Federal Ministry for Education and Research Virtual Liver Network initiative (grant 031 6154 ) and is a recipient of an ORISE Stipend of the FDA.
Y.W. and Z. Lin contributed equally to this work.
J.B. and W.T. contributed equally as senior authors.
The opinions expressed in this article do not reflect the official positions or policies of the US Food and Drug Administration.