Efforts for the development and application of therapeutic cancer peptide vaccines
NEC is engaged in the drug discovery business using AI technology in the medical and healthcare field. In 2016, NEC established a venture company called Cytlimic Inc. in order to promote the clinical development and application of therapeutic cancer peptide vaccines discovered through collaborative research with Yamaguchi University and Kochi University. NEC will proceed with the new cancer drug discovery activities through co-creation with Cytlimic Inc., Yamaguchi University, and Kochi University.
We will ask Professor Udaka (Kochi University) and the person in charge of the HealthTech Business Development Office, NEC, about the background and significance of the co-creation.
Keiko Udaka Professor, Department of Immunology, Kochi Medical School
The main research theme is T-cell recognition. She is engaged in the development of targeted immunotherapies for malignant tumors and the technologies to suppress allergies using identification technologies for MHC binding antigen peptides that present antigens to T cells.
Akira Kitamura Senior Manager
HealthTech Business Development Office, Business Innovation Strategy Department, NEC Corporation
He is in charge of developing new businesses in the drug discovery and related areas.
Tomoya Miyakawa Expert, Ph.D. (Doctor of Pharmacy)
HealthTech Business Development Office, Business Innovation Strategy Department, NEC Corporation
He is in charge of developing technologies in the new drug discovery business.
Peptide vaccines that activate the T cells to attack cancer cells
ーProfessor Udaka (a collaborative research partner of NEC and Cytlimic Inc.), could you give us a summary of your research?
Udaka：I currently research the immune functions of T cells (T- lymphocytes). T cells can clearly tell the difference between normal cells and abnormal cells, such as virus-infected cells, and work to eliminate foreign bodies. T cells have an excellent mechanism where they only eliminate malignant cells and do not damage adjacent healthy cells even though they exist together. Although cancer cells and healthy cells look identical, T cells are able to tell the difference; however, the way T cells make the distinction has only been partially clarified. I would like to identify the markers for malignant cells and develop technology to eliminate cancer cells that should be eliminated but have not yet. I would also like to successfully complete the development of a cancer vaccine in the near future.
ーCould you tell us about why you became a co-creation partner?
Udaka：I came back to Japan in 1994 after researching T cells in the United States and Germany for 10 years. When I was working at the Juntendo University laboratory, I met a researcher who was on a temporary transfer from NEC and was looking for research ideas that could be used in the medical and biotechnology fields employing information technology. Around that time, the mechanism of how peptides*1 bind to major histocompatibility complex (MHC) molecules*2 and how T cells identify amino acid sequences in the peptides to determine whether they are cancer cells had just been revealed. As for amino acid sequences, there are an enormous number of possible combinations (nine each of 20 types of amino acids). I thought that information technology might be used to identify the characteristics of antigens among such combinations; accordingly, the collaborative research on the peptide prediction system with NEC commenced.
ーCould you tell us about the process and the progress of the collaborative research between Professor Udaka and NEC?
Kitamura：The central research laboratory at NEC has carried out basic research in machine learning since 1991. After we met Professor Udaka, we commenced the collaborative research of peptide sequence prediction. In the mid-2000s, biotechnology was on a downward trend, and NEC had to downscale its activities; however, during that period, Professor Udaka continued clinical studies on WT1 peptides (which are the results of collaborative research with NEC). In 2011, Professor Udaka told us that WT1 peptides had shown efficacy for some patients. Accordingly, because the effectiveness of NEC’s peptide prediction technology was reevaluated, the peptide drug discovery project was launched as a new business in 2012. After that, we commenced collaborative research on hepatocellular carcinoma with Yamaguchi University and Kochi University, which led to the establishment of Cytlimic Inc. in 2016. If Professor Udaka had not continued her clinical study, we would not have entered the drug discovery business or established Cytlimic Inc. She saved us.
Udaka：Before the efficacy of the cancer immunotherapy was confirmed, pharmaceutical companies were only observing progress while maintaining a certain distance, and they had even invested in developing vaccines; however, in recent years, cancer drugs, such as Opdivo (*3), which is designed to overcome immune suppression, have been introduced into the market, and we have entered an age where vaccines are viewed positively. It has been found that whether or not the T cells of patients naturally increase is one of the major factors that cause the difference between cases in which the clinical response of Opdivo is shown to be positive or negative. Accordingly, even though Opdivo had no effect on patients, there is a strong possibility that it can be more effective if T cells, which recognize and kill tumor cells, increase through the active immunization with peptide vaccines. Thus, the efficacy of peptide vaccines has been re-appreciated; accordingly, the pharmaceutical industry has become more positive towards the strategic development of peptide vaccines.
Predicting results from approximately 500 billion amino acid sequences using machine learning
ーWhat are the advantages of collaborative research with NEC compared to research carried out by other universities?
Udaka：The major advantage is that we can efficiently predict the ability of peptide binding to HLA molecules for random peptide combinations of amino acid sequences using information technology. There are twenty kinds of amino acids required for human life; the length of a peptide recognized by killer T cells consists of nine amino acids, and the length of a peptide recognized by helper T cells that activate killer T cells consist of eleven or more amino acids. In the case of nine amino acids, there are approximately 500 billion combinations of amino acid sequences. It is hard to search for peptides that bind to HLA molecules from among these combinations. For instance, there was an experimental method to identify the one and only combination; however, in the case of HLA-binding peptides, around one in one hundred peptides from among approximately 500 billion types binds to HLA. Thus, it is a rough identification of combinations. As there is no way to even estimate the combinations, it is more efficient to use machine learning to identify the combinations.
ーWho did NEC need as a partner when working on the drug discovery business?
Miyakawa：I joined NEC in 2002 after working for a pharmaceutical company and have worked to develop new business in the drug discovery field using the IT technologies of NEC. Obviously, the partnerships with experts who possess broad knowledge and experience in the field from basic research to clinical development are essential on the co-creation stages. It is also essential to find those who are interested in the use of IT and can understand the technologies of NEC.
Kitamura：Completely different approaches (stances) are required for experiments involving industrial products and biological experiments. As for industrial products, identical products can be manufactured as many times as needed; however, for biological experiments, the results may often be different depending on the techniques of experimenters even for the same experiment. Accordingly, our drug discovery requires expertise and high-quality experience in experiments. Therefore, the significant factor was the encounter between our machine learning technologies and an expert like Professor Udaka.
ーHow did you utilize machine learning?
Udaka：My research consists of countless repetitions of the same tasks. I conducted the experiments on the binding of peptides over and over again. It is important to obtain accurate numerical data in the experiments. I am particular about obtaining more accurate numerical data than anyone else. I have so far measured the binding of around 1000 peptides. At some point, I synthesized around 500 per year. As for HLA-binding peptides, even overseas databases did not contain such a large volume of accurate information.
Miyakawa：Although data is important for machine learning, many overseas databases are simply a collection of data from the literature where peptide-binding was measured by various methods with limited accuracies. However, Professor Udaka provided stable, high-quality data. Accordingly, we successfully conducted machine learning based on a reliable dataset.
Udaka：During the first joint research, we repeated machine learning where peptide binding to HLA molecules was predicted for 100,000 amino acid sequences of peptides selected from the random number table. As a result, some sequences could be predicted while other sequences could not be predicted. The reason why some sequences could not be predicted was the lack of information on the binding. Accordingly, I measured binding for the sequences that were most difficult to predict, and then entered the data into the database. As a result, learning progressed dramatically, and it is now possible to predict peptide binding to HLA molecules for random peptide combinations of amino acid sequences. At that time, the accuracy of the prediction using overseas public databases was around 30% to 40%, and the accuracy was around 50% to 60% even when our peptide library method was used. However, the accuracy of NEC’s prediction system was 93%. Thus, we successfully made a world-leading achievement.
Miyakawa：By using our current peptide predication system, we can identify peptides within a second. In order to reach this stage, it took around a year for us to establish the rules and alternately repeat machine learning and the professor’s experiments for one Human Leukocyte Antigen (HLA) type. We are currently considering the expansion of the range of prediction to different types of HLA molecules with Professor Udaka.
Provision of new treatment based on the result of the clinical study
ーWhat have you accomplished so far?
Udaka：I have carried out clinical studies for 10 years. In terms of prostate cancer, positive results, such as the reduction in tumor markers, the reduction in tumor (cancer) growth, and the suppression of cancer progression, have already been produced in around 40% of all cases through the administration of an HLA-binding peptide (which are served as a marker for cancer cells) as an immunogen. In the development process, there are many potential peptides that can be used for immunization, and the use of only one type of peptide can immunize most Japanese people against cancers if we use rare peptides that bind to some of the major types of HLA*2 that Japanese people possess. Therefore, I plan to use such rare peptides for the development of the next cancer vaccine. In addition, positive results have also been produced for brain tumors. Moreover, I have worked on cancers in the head and neck and sarcomas and bone malignancies in the field of orthopedic surgery. WT1 (cancer antigen), which is currently being targeted, is expressed in approximately 70% to 80% of solid tumors and in leukemia. I would like to complete the preparation of the clinical study of the next-generation vaccines for prostate cancer by the end of this fiscal year.
ーCould you tell us about the future prospects?
Kitamura：Cytlimic Inc. plans to complete the preparation of the clinical study within two years after the establishment and a Phase II trial within three years after completion of preparation. It is an important objective for NEC, the shareholder of Cytlimic Inc. Professor Udaka also serves as a scientific advisor for Cytlimic Inc. Moreover, NEC intends to launch new efforts in the area of immunotherapy while keeping pace with Cytlimic Inc.
Udaka：I would like to deliver the next-generation peptide vaccines as soon as possible. Patients with prostate cancer do not have effective treatments if hormone therapy does not work. Therefore, clinicians expect me to develop new treatments for patients requiring new drugs, and I would like to work hard to meet their expectations.
ーWhat do you expect of each other as a co-creation partner?
Udaka：First, we would like to develop peptide vaccines for cancers and chronic viral diseases, as well as peptide immunotherapies to treat allergies and autoimmune diseases. Current technologies provide only the methods to enhance or suppress immune responses in an antigen independent fashion. What is sought in the future is a method to control immune responses in an antigen-specific manner. For instance, a method to suppress harmful immune responses observed in allergies and autoimmune diseases while maintaining the protective immunity towards pathogenic microorganisms. Once we obtain commands on identifying peptide antigens responsible for the disease, it would then be possible to develop a method to control immune responses positively or negatively only on those peptides. As a future prospect, we would also like to cooperate with each other to enrich the medical information resources in Japan. Once genetic information on HLA types is obtained, it can be used for the treatment of infectious diseases and cancers, as well as allergies and autoimmune diseases. If we have rich medical information resources, it would also enable the HLA matching required for transplantation in a short time. Under the current system, if individuals register as donors, their genotypes and HLA types are identified, and such data is stored in the transplant database; however, the data cannot be used for treatments, such as peptide vaccines even if donors develop diseases. If registered donors develop diseases, their genetic information has to be identified again (which require expenses again) because of reliability problems in self-reporting the HLA types by patients. If the genetic information stored on the network becomes available, the cost can be reduced, and the quality and speed of medical services provided are enhanced. Therefore, I think we should have a genetic information database system that is useful for medical use and can be accessed by both physicians and patients. I would like to work together with NEC to consider the creation of such a system.
Kitamura：We would like to ask Professor Udaka for her continued support in the launch of NEC’s AI-based drug discovery business in terms of in-depth knowledge about immunology, personal connections, and experiments. Moreover, we would like to jointly consider the efforts for the further use of medical information.