AMED-funded projects

Project summaries 2019

Prying open the black box of deep learning

Artificial intelligence can predict the probability of prostate reoccurring in a way that human doctors can glean insights from

An automatically annotated 3D whole-mount pathology image. The deep-learning algorithm automatically discovers features from diagnostic annotation-free histopathology images and presents them in an understandable way. Areas with a high probability of cancer recurrence are shown by height and color in the 3D pathology image.

© 2020 AMED

The enormous potential of artificial intelligence (AI) is just beginning to be unleashed in medical research and clinics. In some instances, AI can even outperform human experts, but an ongoing problem is that its process is opaque to humans.

Now, an AMED-funded study led by Yoichiro Yamamoto of the RIKEN Center for Advanced Intelligence Project is lifting the lid on the black box of AI. Yamamoto’s team developed a deep-learning system that predicts the probability that prostate cancer will reoccur based on pathology images from patients.

The AI results were more accurate than those of human pathologists — human pathologists scored an accuracy of 0.744 (area under the curve) on a scale between 0 and 1 (where a score of 1 indicates perfect accuracy), while the machine achieved an impressive 0.820.

But more importantly, the deep-learning system generated images that highlighted the regions that had the highest AI-assigned weightings, thus enabling human researchers to glean new knowledge. “Most AI techniques have only been used for classification, so they essentially mimic human pathologists,” explains Yamamoto. “But our deep-learning technology goes beyond that — it can glean new insights.”

One example of a novel insight that the team’s deep-learning technology has uncovered is that it sometimes highlights images that don’t include any cancer cells. “AI says these images are very important for predicting cancer recurrence, but you won’t find them in pathology books produced by humans,” says Yamamoto. “This is new knowledge.”

The technique is not restricted to prostate cancer. Yamamoto and his team are now applying it to other cancers including rare cancers.

Project: Project for Research on Creation of Medical Arts

Researcher: Yoichiro Yamamoto

Host Institution: RIKEN

AMED Funding: ICT Infrastructure for the Establishment and Implementation of Artificial Intelligence for Clinical and Medical Research

In sight of a stem-cell therapy for corneal disease

Eye tissue made from ‘reprogrammed’ stem cells restores vision in world’s first recipients

A false-color transmission electron micrograph showing the typical corneal stroma, the thickest layer of the cornea. A new stem-cell therapy could help repair damaged cornea.


The first person in the world to receive a corneal transplant made from induced pluripotent stem cells (iPSC) continues to show signs of improved vision with no serious complications more than six months after undergoing the procedure.

Donor cells from a non-ocular source were reprogrammed into an undifferentiated state and expanded in the laboratory to form a multi-zone of eye-like cells called a self-formed ectodermal autonomous multi-zone (SEAM). Cells from the SEAM were then coaxed to form the type of stem cells found in the cornea, a transparent layer that covers and protects the eye. A thin sheet of the stem cell–derived corneal cells was then laid across the top of their eyes. This represents a major milestone for regenerative medicine — and has already led to a second patient receiving the same treatment.

“For these two patients, the outcome after surgery has been excellent, both in terms of safety and efficacy,” says Kohji Nishida, an ophthalmologist at the Graduate School of Medicine, Osaka University, who is leading the AMED-sponsored trial.

“In addition, both of the transplanted iPSC-derived cell sheets showed good quality in terms of purity, cell viability, cell number, marker expression, sterility and genome stability,” says Osaka University tissue engineer Ryuhei Hayashi, who is handling the cell preparation.

Nishida and Hayashi expect to administer their experimental regenerative therapy to two more people before the end of 2020.

If the new treatment lives up to its early promise, it could become a practical and scalable way to correct vision loss for the 12.7 million people worldwide who are living with corneal blindness as a consequence of damage to the eye’s outer casing. Currently, patients must wait for corneal tissue to become available from deceased donors. But with only around 200,000 corneal transplants performed globally each year, less than 2% of affected individuals receive the sight-restoring treatment they need.

Project: Japan Regenerative Medicine Project

Researcher: Kohji Nishida

Host Institution: Graduate School of Medicine, Osaka University

AMED Funding: Research Center Network Program for Realization of Regenerative Medicine
Research Project for Practical Applications of Regenerative Medicine

Fending off new forms of flu

The lining of our lungs provide an essential safeguard against influenza subtypes that successfully infect other species

The difference between influenza viruses that successfully infect both birds and mammals — like the H3N2 virus shown here — and those that remain in birds likely comes down to a prompt immune response.

© CDC/Science Photo Library/Getty

AMED-funded researchers are exploring why certain subtypes of influenza pose a dire threat to humanity while others remain restricted to non-mammalian hosts such as birds. The team has identified one layer of protection against infection.

The outcome of influenza exposure is determined in large part by the initial immune response. A strong inflammatory reaction may eradicate the virus quickly, averting a long battle with sickness that could have serious or even fatal consequences. The inflammatory response to influenza is mainly governed by immune cells known as macrophages, but the epithelial cells that form the lining of the lungs also play a role in this process.

With funding from AMED as part of the Research Program on Emerging and Re-emerging Infectious Diseases, Atsushi Kawaguchi and colleagues performed a screen on these cells to identify genes associated with inflammatory response in mice. They homed in on a gene encoding a protein called MxA.

“MxA has long been known to be a determinant of pathogenicity for virus infection, and the host range of avian influenza,” says Kawaguchi.

His team has now demonstrated that MxA directly interacts with influenza proteins, triggering a rapid and potent inflammatory response against infection. These findings could explain why humans fend off many strains that infect bird species, highlighting a critical line of defense that stops influenza strains from jumping hosts to cause outbreaks.

Kawaguchi now hopes to dissect this viral protein recognition more carefully to learn why epithelial cells generate such a strong inflammatory response to certain viruses.

Project: Molecular basis for mammalian adaptation of emerging influenza virus

Researcher: Atsushi Kawaguchi

Host Institution: University of Tsukuba

AMED Funding: Research Program on Emerging and Re-emerging Infectious Diseases

Split fear processing in the brain

Different parts of the brain are associated with the emotional and physiological responses to fear and anxiety

An image of the brain showing the relative importance of different brain regions to the prediction of subjective fear. Darker regions (both blue and red) have higher predication weights than lighter ones (red indicates a positive weight and blue a negative weight).

Reproduced from Taschereau-Dumouchel, V., Kawato, M. & Lau, H. Multivoxel pattern analysis reveals dissociations between subjective fear and its physiological correlates. Mol. Psychiatry (2019) doi: 10.1038/s41380-019-0520-3 and licensed under CC BY 4.0. © 2016 V. Taschereau-Dumouchel et al.

Bodily manifestations of distress and alarm are associated with different parts of the brain than those associated with the psychological states related to these emotions, an AMED-funded study has found.

In studies of fear, anxiety and trauma-related disorders, researchers have long relied on physiological responses to threat, such as increased sweating, as a proxy for psychological states. Consequently, drugs that manage outward signs of fear or anxiety may not actually help with the subjective suffering that typically underpins most neuropsychiatric disorders.

However, “by paying more attention to the subjective report of patients, we might be able to target new components of fear and anxiety disorders that may have been overlooked by approaches focusing exclusively on physiological responses,” says Vincent Taschereau-Dumouchel, a cognitive neuroscientist at the ATR Computational Neuroscience Laboratories in Kyoto and the University of California, Los Angeles, who co-led the study.

Taschereau-Dumouchel and his colleagues trained an artificial-intelligence algorithm to read scan data from functional magnetic resonance imaging of the brain and then differentiate between objective physiological reactivity to fear (as measured by skin conductance) and subjective fear reports.

They presented people with images of cockroaches, spiders and other scary creatures, and found that certain brain regions, such as the amygdala and insula, were primarily behind physiological responses to fear, whereas other regions involved in metacognition and conscious perception played a bigger role in subjective experiences.

The researchers suggest that a greater focus on self-reported measures of fear — and the brain regions involved — could help optimize the treatment and management of fear and anxiety disorders.

Project: Psychiatric & Neurological Disorders

Researcher: Vincent Taschereau-Dumouchel

Host Institution: ATR Computational Neuroscience Laboratories

AMED Funding: SRPBS, Brain/MINDS

Cancer–nerve crosstalk

Therapeutically targeting the nervous system could halt the spread of breast tumors

AMED-funded research has found that simulation of the sympathetic nervous system hastens the growth of breast cancer.


Nerves play a critical role in the growth of breast tumors, a finding that could lead to new ways of predicting the aggressiveness of breast cancer and to novel strategies for treating the disease.

A gene therapy technique to locally and selectively manipulate the neural input of tumors in rodent models of breast cancer has been developed by an AMED-funded team led by molecular physiologist Atsunori Kamiya from Okayama University in Japan has developed

In both rats and mice, cancer progression sped up following stimulation of sympathetic nerves, which normally direct the body’s fight-or-flight response to dangerous or stressful situations, but slowed down after activation of parasympathetic nerves, which counteract sympathetic signals. Targeted ablation of tumor-specific sympathetic nerves similarly curbed tumor growth and reduced the expression of surface molecules that suppress anti-cancer immunity.

The gene therapy approach also caused greater tumor inhibition than β blockers, a class of drugs that impede sympathetic nerve signaling and have been found to restrict the proliferation of breast cancer.

Clinical validation for the rodent data came from an analysis of tumor samples from 29 women with breast cancer. The researchers showed that patients with higher densities of sympathetic nerves in their tumors had more indicators of a restrained immune response and were more likely to suffer early recurrence of cancer.

The study — a part of the AMED’s Japan Cancer Research Project and the Project for Psychiatric and Neurological Disorders — suggests that controlling anxiety, anger and other sources of stress that arouse the sympathetic nerve system could help improve outcomes for patients with breast cancer, Kamiya says. It also highlights a potential way to “selectively manipulate local neural input for therapeutic purposes — a technique that has been recently termed cancer neural therapy,” he adds.

Project: Japan Cancer Research Project, Project for Psychiatric and Neurological Disorders

Researcher: Atsunori Kamiya

Host Institution: Okayama University


Brain imaging reveals atomic structure of Parkinson’s brain clumps

An AMED-funded study supports the idea that pathogenic protein aggregates can propagate through the brain

The red specks in this fluorescent micrograph of a section through a brain affected by Parkinson’s disease are the protein alpha-synuclein, which is thought to cause the progressive degeneration of the neurons that leads to the symptoms of Parkinson’s disease. Researchers funded by AMED have now observed the structure of alpha-synuclein aggregates.


The distinctive structure of abnormal protein aggregates in the brains of people with Parkinson’s disease has been observed for the first time by researchers from Osaka University Graduate School of Medicine and the Japan Synchrotron Radiation Research Institute. This achievement has important implications for diagnosing and treating the devastating disease.

Using state-of-the-art imaging techniques on autopsy brain tissue, the AMED-backed team showed that the clumps of α-synuclein that form inside the nerves of a Parkinson’s-affected brain actually adopt an ordered structure consisting of long fibers arranged into sheets.

Others had shown that α-synuclein could fashion itself into this shape in a lab dish and that, when implanted into the brains of mice, convert the folding of other α-synuclein proteins. However, there was no prior evidence of this phenomenon in human patients, and the idea of α-synuclein behaving like a prion — a type of misfolded protein that transmits its shape in an infectious, wave-like manner — remains controversial among Parkinson’s researchers.

Given the structure observed in the autopsy brains, “this could be important evidence to support the prion hypothesis in Parkinson’s disease,” says Hideki Mochizuki, a neurologist at Osaka University, who led the study.

In other AMED-supported work, Mochizuki and his colleagues have identified ways of rapidly diagnosing α-synuclein aggregates in cerebrospinal fluid and then targeting the toxic protein with an antisense drug. The latest imaging data, he says, “strongly support” the idea of using these technologies to identify patients with early signs of α-synuclein build-up and then treating those individuals early to prevent further transmission of the disease process.

Project: Psychiatric & Neurological Disorders

Researcher: Hideki Mochizuki

Host Institution: Osaka University Graduate School of Medicine


A new way to find cancer-suppressing cells

Cells that express a protein called Meflin help keep pancreatic tumors in check

Fluorescence light micrograph of human pancreatic cancer cells obtained in the laboratory. An AMED-funded study has found that the fibroblasts (green) surrounding pancreatic tumors (red) that express a protein called Meflin help keep the cancer in check (nuclei in blue).

© 2020 Atsushi Enomoto

A way to identify cells expressing a protein that curtails the growth of pancreatic cancer has been found by researchers supported by AMED’s Japan Cancer Research Project.

Part of what makes cancer so deadly is the helping hand that tumors receive from neighboring cells known as cancer-associated fibroblasts. But not all such cells serve as abettors of tumor progression — a subpopulation of these fibroblasts restrains, rather than promotes, cancer growth.

Looking at the cells surrounding pancreatic tumors, a team led by Atsushi Enomoto and Masahide Takahashi from the Nagoya University Graduate School of Medicine discovered that a protein called Meflin is expressed on the surface of cancer-restraining fibroblasts.

In mouse models, Meflin deficiency produced more-aggressive tumors, while patients with more Meflin-expressing fibroblasts infiltrating pancreatic tumors generally had better survival rates.

Diagnostic tests that evaluate levels of Meflin in the tumor microenvironment could thus help doctors predict the likely course of a patient’s disease and develop appropriate management strategies, says Atsushi Enomoto, an associate professor at the Nagoya University Graduate School of Medicine who co-led the study. “A therapy that specifically controls Meflin expression could be valuable,” he adds.

Enomoto and his team also showed that genetically enhancing Meflin levels helped suppress the growth of tumors implanted in mice. Vitamin D is one molecule known to boost expression levels of Meflin, so the findings provide a biological rationale for combination trials involving vitamin D analogues for pancreatic cancer, Enomoto notes.

Clinical studies are also ongoing involving vitamin D supplementation for tumors of the breast, prostate and other tissues. “But for now, the presence of Meflin-positive cancer-restraining fibroblasts has been shown only in the context of pancreatic cancer,” Enomoto says. “We need to look at other cancers in the future.”

Project: Japan Cancer Research Project

Researcher: Atsushi Enomoto

Host Institution: Nagoya University Graduate School of Medicine


The dangers of a discriminating diet

Evolution shapes the metabolic response to different foods and the extent to which organisms benefit from their consumption

By studying how fruit flies have adapted in response to available food sources, researchers hope to gain insights into how human genetics and dietary choices combine to shape our risk of metabolic disease.

© 2020 Kaori Watanabe

AMED-funded researchers are examining the evolutionary factors that underlie the flexibility to adapt diet.

Some people are simply pickier about what they eat than others. But dietary decisions have also influenced our evolution, shaping the extent to which we can draw nourishment from different types of food.

With support from AMED, researchers led by Yukako Hattori, Kaori Watanabe and Tadashi Uemura are examining the evolutionary factors behind this.

“Some species, including humans, are nutritional generalists, which have adapted to a wide range of food resources, while others are specialists that feed on specific resources,” explains Hattori. “But much remains to be discovered about how generalist species respond and adapt to diversified nutritional environments.”

Hattori, Watanabe and Uemura recently explored this question by looking at two different fly species: generalist Drosophila melanogaster and its specialist cousin Drosophila sechellia, which has evolved to feed on one particular type of plant. They fed larvae of both species different diets and monitored differences in gene expression, metabolism and growth.

The results revealed striking differences in their response to carbohydrate-rich diets — D. melanogaster exhibited regulatory activity in hundreds of metabolic genes and flourished on such diets, while D. sechellia lacked this coordinated response to carbohydrates and fared much more poorly.

D. sechellia may have lost this system under their constant low-carbohydrate diet,” says Hattori.

This highlights an important role for environmental factors in shaping dietary diversity. Hattori notes that we share a great deal of our metabolic machinery with the humble fruit fly, and her team ultimately aims to pursue comparative studies that explore how the interplay between our diet and our genetic background might contribute to metabolic disorders like diabetes and obesity.

Project: Clarification of the mechanism of individuals’ functional impairment over the entire life course

Researcher: Yukako Hattori (primary AMED grant recipient, Tadashi Uemura)

Host Institution: Kyoto University

AMED Funding: AMED-CREST, Clarification of the Mechanism of Individual’s Functional Impairment over the Entire Life Course

Return to AMED-funded projects

If you are interested in reviewing AMED grant applications and becoming one of the first to learn about the world-leading medical research coming from Japan, please sign up by clicking the button below.