The LLNA: BrdU-ELISA can be used to classify GHS category 1 chemicals, but a sub-categorization criterion has not been provided. Here, we reanalyzed 32 sensitizers listed in the ICCVAM peer-review data set and proposed an optimal sub-categorization criterion for LLNA: BrdU-ELISA as 6% when using EC1.6. Additionally, we confirmed its applicability to CBA/J mice with 15 sensitizers listed in the LLNA performance standard and confirmed correct assignment of their sub-categories. Consequently, the newly proposed criterion appears to be applicable for practical use for GHS sub-categorization.

To develop read-across assessment for chemicals with the potential of inducing hemolytic anemia, we took the category approach based on similarities of the chemical structure and mode of action. Firstly, an integrated database was constructed by collecting repeated dose toxicity datasets and metabolism information from Hazard Evaluation Support System Integrated Platform (HESS) in addition to various toxicology databases including ToxRef, COSMOS, and RepDose. In the database, approximately 447 of 1550 substances were identified as those of hematotoxicity that induced hemolytic anemia. Secondly, the category approach based on chemical structures and toxicity mechanisms of the hematotoxic substances led to the following categories: anilines (Category 1), nitrobenzenes (Category 2), nitroanilines (Category 3), hydrazines (Category 4), oximes (Category 5), ethylene glycol and alkyl ethers (Category 6). The mechanisms of toxicity were postulated that: reactive metabolites of substances in Categories 1 to 3 oxidize hemoglobin; substances in Categories 4 and 5 directly oxidize hemoglobin; and reactive metabolites of Category 6 substances modify erythrocyte membrane. Toxic mechanisms of the substances in these categories were directly linked to hemolysis, whereas hematotoxicity induced by a number of substances unclassified to the categories may have indirectly occurred. These categories that are helpful to further clarify structural boundaries associated with hematotoxicity and contribute to constructing AOPs should support read-across assessment for predictive toxicology.

In the Risk Assessmen-(Ⅰ) for human health under the Chemical Substances Control Law, carcinogenicity should be evaluated quantitively if possible. In the present study, we tried to evaluate carcinogenicity of some mutagenic priority assessment chemicals (PACs) quantitatively. Using slope factor, unit risk or TD₅₀ for each chemical or a value of threshold of toxicological concern for PACs without sufficient carcinogenic information, it was clarified that risk assessment for all the mutagenic PACs could be prioritized.

In the early stages of drug development, prediction of pharmacokinetic profiles of new chemical entities is essential to minimize the risks of potential withdrawals, and computer-aided drug design that predicts ADMET parameters using in silico models has recently attracted attention. In this study, we will introduce predictive models for protein binding and volume of distribution of drugs in humans and brain penetration considering the efflux ratio of P-glycoprotein, which are necessary for integrated pharmacokinetic analysis.

Quantitative and objective evaluation is required for clinical observation in general toxicity studies, but actually, it depends on the subjectivity and experience of observers. In this study, we verified whether higher quality clinical observation is possible by incorporating sensing technology. The accelerometers were attached non-invasively to the dogs, and we tried to detect changes in the amount of activity and the specific behaviors using accelerometry data. As a result, it was suggested that quantitative, objective and continuous evaluation is possible.

In this study, we focused on the developing in silico model to predict a Drug-Induced Liver Injury（DILI） based on the mechanism of toxicity. Using the drug-induced liver injury rank（DILIrank）dataset DILI classification model was built with XGBoost as a machine learning algorithm from chemical structure information and in vitro test information. These models showed 0.78 as the average ROC-AUC from five times of the nested cross-validation. These results suggest that we could constructed DILI prediction models with considering the mechanism of DILI toxicity.

In recent years, the use of electronic data compliant with SEND, a standard for the exchange of non-clinical study data, has driven toxicologists and data scientists to visualize the data. However, it is not easy for a data scientist to achieve the visualization and analysis desired by a toxicologist.

We have discussed the key points to visualize and analyze the data from different standpoints of toxicologists and data scientists, and made them document as a common understanding. In this presentation, we will provide our knowledge on how to effectively use SEND for visualization.

The number of study types in scope of SEND has gradually increased for submissions to the U.S. FDA, and “studies related to embryo-fetal development” are the next (DART studies starting on and after March 15, 2023 for NDAs and those starting on and after March 15, 2024 for INDs), and the guideline of SENDIG DART v1.1 had already been finalized by CDISC. In the SENDIG DART v1.1, new 7 domains were added and modifications specific to DART studies were made onto the existing domains. It seems difficult to be compatible with these alterations only with the knowledge of SEIDIG v3.1, and therefore understanding of SENDIG DART v1.1 is essential. Under this circumstance, we have created SEND data in accordance with the SEND DART v1.1 using in-house study data on embryo-fetal development in rabbits to investigate issues that may occur during creation of SEND data for DART studies. As a result, we encountered various issues such as information that cannot be interpreted from the guideline, examinations that are not included in the controlled terminologies, SEND conversion of observation methods specific to the test facility and methodology of data population for animals that aborted. Despite such difficulties, we managed to complete the creation of SEND dataset through discussions with the involved parties including our business associate. In this presentation, we summarized the specific issues occurred during the SEND dataset creation and practical examples of the methodologies we had established through the discussions.

Based on "Draft Advisory Document on GLP Data Integrity" (OECD GLP Working Group) and "Items to be implemented by the sponsor with respect to external contracts archiving facility" (PMDA, 25th GLP training workshop. 2013), we investigated that how the sponsor should assure the quality of the electronic data using the Cloud to comply with GLP requirements in Japan. In this report, referring to the operation of overseas data centers using SaaS systems, we will explain the points that sponsors should check regarding the electronic data management in the Cloud to ensure the reliability of GLPs.

Due to the mandatory submission of SEND (Standard for Exchange of Nonclinical Data) data by the US FDA, many pharmaceutical manufacturers and CROs are increasing their experiences in creating SEND data.

Under this circumstance, for our company or a CRO which does not currently own a SEND data creation system, it is necessary to build a SEND system in cooperation with a partner company.

For this reason, we prepare SEND data of the study conducted at our facility by collaborating with Ina Research Inc., a member of G-SEND (Global SEND Alliance), which is a consortium aiming for uniform and high quality of SEND data between facilities.

When multiple facilities collaborate in creating SEND data, it is a critical issue how to cooperate among the facilities and allocate the operations because it affects the schedule, work (cost) load, and scope of responsibility. Therefore, we used a model study for SEND to prepare SEND data in collaboration with both companies, and examined the optimal work allocation and a scope of responsibility in order to create high-quality SEND data efficiently.

In this presentation, we will summarize the issues and countermeasures when multiple facilities collaborate to create SEND data.

In this study, we sought to develop in silico models for predicting the inhibitory activity of chemical substances against cytochrome P450 (CYP) based on chemical structure information. We used in vitro experimental values of 326 substances for rat CYPs and 215 substances for human CYPs from the HESS as learning data. The models established by XGboost showed ROC-AUC of 0.8 or more for rat and 0.75 or more for human. In this study, we have developed high-performance and reliable classification models to predict the inhibitory activity of chemical substances against CYPs by in silico methods.

We would like to introduce the automatic quality check (QC) workflow for numerical SEND data, obligated to submit to FDA, implemented by open source platform KNIME and the improvement for time-consuming QC tasks. We can finish QC for all numerical data in several tens of seconds using KNIME workflow and obtain QC results as Excel files with highlighting improper matching points. Since comparison data source is the text file converted as general-purpose un-pivot format from the Appendix, our QC workflow is applicable to any data from various types of toxicity test systems.

Standard for Exchange of Nonclinical Data (SEND) was published as the data standard format for nonclinical studies by CDISC.

Applicants to FDA must submit the electronic study data of nonclinical safety studies (SEND dataset), not only in compliance with CDISC standard (SEND Implementation Guide, Controlled Terminology, etc), but also with the FDA-issued documents (Study Data Technical Conformance Guide, Business Rules, etc.).

CJUG SEND team created a check tool to check SEND dataset with SAS program based on the check items covered the above documents and rules. We will show the usefulness of check tool to ensure the quality of SEND dataset.

Recent developments in deep learning will have a significant impact on the field of toxicology. In this study, we attempted to construct an explanatory AI model for discriminating genotoxicity (AMES) from the structure of compounds using kGCN Kojima and Okuno developed. We constructed a dataset by collecting three types of public AMES data and assembled them into three data sets. By applying kGCN to the three data sets, we obtained models with high accuracy. The results of this analysis will be reported together with the visualized compound structures.

Acknowledgment: This work is supported by AMED (DAIIA) under Grant Number: 21nk0101111h0102.

Large amounts of data associated with safety issues are generated along the entire lifetime of drugs, from its infancy as preclinical leads, through its adolescence as clinical candidates, all the way up to its adulthood as marketed drugs exposed to the human population. Across the different stages in the life of a drug, some of the data collected initially may be confirmed and consolidated with data at an advanced stage, whereas other data may not be translated, and in some cases may even contradict, those safety signals that are ultimately observed in the human population. Collecting and properly integrating such an heterogenous pool of data is already a challenge in itself. But even if one manages to put all data together, the analysis is not straightforward, and it may require the assistance of purposely designed visual analytics tools.

To this aim, we introduce the newly designed Translational Safety Charts that collect, in a single image, data from safety pharmacology, preclinical toxicology, clinical safety and postmarketing surveillance. They are implemented in CLARITY.PV, a new pharmacovigilance web-accessible platform that currently contains Translational Safety Charts for 8,060 drugs. A thoroughly validated methodology has been implemented to detect consolidated drug safety signals and link them to pharmacology, preclinical and clinical biomarkers. The analysis of safety signals can be extended beyond individual drugs to drug classes and stratified by age, weight, gender and geographical location.

Liver Malignant Tumors (LMT) have been reported as adverse drug reactions in clinical settings. However, it is difficult to conduct comprehensive in vivo studies to evaluate factors of drug-induced LMT. Therefore, we attempted to define signals of LMT-inducing drugs by multivariate analysis of FAERS and to construct a prediction model for LMT signal detection using mutagenicity, and agonist/antagonist activities against nuclear receptors and stress response pathways. We consider examining the model to estimate various information of Adverse Outcome Pathways related to drug-induced LMT.

The nSDRG is a document to explain the SEND datasets. PhUSE guideline and template are very informative in preparing the nSDRG; but it is not easy to understand the details. The nSDRG subteam of CJUG SEND team picked up the issues in preparing nSDRG and shared them with PhUSE nSDRG working group, as introduced in the last JSOT meeting in 2020. We discussed the solution of the issues with referring the draft guideline published in Jan. 2020, in which some parts were revised considering our concerns. We will provide some approach, although it is still under review by the PhUSE.

Colorectal cancer (CRC) organoids and paired cancer-associated fibroblasts (CAFs) from surgical specimens were established and we evaluated gene expression profiles of organoids with and without co-culture with CAFs. Original CRC showed variable expression profile for intestinal stem cell marker genes, e.g. LGR5, ORFM4. These characteristic gene expression patterns were maintained in each organoid model. On the other hand, we found that expression levels of several genes, which are substantially expressed in original CRC, were downregulated in organoids but reproduced by co-culturing with CAFs. They comprised immune response- and external stimulus-related genes, e.g.,REG family and dual oxidases (DUOXs), which are known to have malignant functions, leading tumor cells to proliferative and/or anti-apoptotic states and drug resistant phenotypes. In addition, the degree of differential induction of REG1 and DUOX2 in the co-culture system varied depending on CAFs from each CRC case.

In conclusion, CRC organoids co-cultured with CAF should be applied for basic and translational researches under partially reproduced tumor-microenvironment.

【Backgrounds】

Bladder cancer (BC) is a kind of malignant tumor derived from the mucous membrane of the bladder. Moreover, it occurs frequently and is prone to recurrence. In especially, muscle-invasive BC progresses quickly and has a high risk of metastasis. However, there are few suitable experimental model cells. Organoid culture method has been applied to tumor research in recent years. In the previous study, we established a novel muscle-invasive BC organoid culture model using the urine samples from BC diseased dogs. Therefore, we focused on generating the bladder lavage fluid-derived BC organoids from human BC patients.

【Materials and methods】

To generate human BC organoids, we collected the cells from bladder lavage fluid of human BC patients and tried to culture them. The composition of the culture medium was changed to more optional components.

【Results and discussions】

The organoids created from bladder lavage fluid has no bacterial contamination and showed stable proliferation. We observed several bladder tissue marker expressions in the organoids and evaluated the difference in cell viability for the anti-cancer drugs. Also, we compared histopathological features, expression of several markers, and gene expression in BC tissues, BC tissue-derived organoids, and bladder lavage fluid-derived organoids. In conclusions, the organoids created from the bladder lavage fluid will be able to be taken advantage of as a valuable experiment model for human BC patients.