Artificial intelligence (AI) has become a relatively common thing; chances are you have a smartphone with an AI assistant or you use a search engine powered by AI. While it’s a broad term that can include many different ways to essentially process information and sometimes make decisions, AI systems are often built using artificial neural networks (ANN) analogous to those of the brain. And like the brain, ANNs can sometimes get confused, either by accident or by the deliberate actions of a third party. Think of something like an optical illusion — it might make you feel like you are looking at one thing when you are really looking at another.

The difference between things that confuse an ANN and things that might confuse us, however, is that some visual input could appear perfectly normal, or at least might be understandable to us, but may nevertheless be interpreted as something completely different by an ANN. A trivial example might be an image-classifying system mistaking a cat for a dog, but a more serious example could be a driverless car mistaking a stop signal for a right-of-way sign. And it’s not just the already controversial example of driverless cars; there are medical diagnostic systems, and many other sensitive applications that take inputs and inform, or even make, decisions that can affect people.

As inputs aren’t necessarily visual, it’s not always easy to analyze why a system might have made a mistake at a glance. Attackers trying to disrupt a system based on ANNs can take advantage of this, subtly altering an anticipated input pattern so that it will be misinterpreted, and the system will behave wrongly, perhaps even problematically. There are some defense techniques for attacks like these, but they have limitations. Recent graduate Jumpei Ukita and Professor Kenichi Ohki from the Department of Physiology at the University of Tokyo Graduate School of Medicine devised and tested a new way to improve ANN defense.

“Neural networks typically comprise layers of virtual neurons. The first layers will often be responsible for analyzing inputs by identifying the elements that correspond to a certain input,” said Ohki. “An attacker might supply an image with artifacts that trick the network into misclassifying it. A typical defense for such an attack might be to deliberately introduce some noise into this first layer. This sounds counterintuitive that it might help, but by doing so, it allows for greater adaptations to a visual scene or other set of inputs. However, this method is not always so effective and we thought we could improve the matter by looking beyond the input layer to further inside the network.”

Ukita and Ohki aren’t just computer scientists. They have also studied the human brain, and this inspired them to use a phenomenon they knew about there in an ANN. This was to add noise not only to the input layer, but to deeper layers as well. This is typically avoided as it’s feared that it will impact the effectiveness of the network under normal conditions. But the duo found this not to be the case, and instead the noise promoted greater adaptability in their test ANN, which reduced its susceptibility to simulated adversarial attacks.

“Our first step was to devise a hypothetical method of attack that strikes deeper than the input layer. Such an attack would need to withstand the resilience of a network with a standard noise defense on its input layer. We call these feature-space adversarial examples,” said Ukita. “These attacks work by supplying an input intentionally far from, rather than near to, the input that an ANN can correctly classify. But the trick is to present subtly misleading artifacts to the deeper layers instead. Once we demonstrated the danger from such an attack, we injected random noise into the deeper hidden layers of the network to boost their adaptability and therefore defensive capability. We are happy to report it works.”

While the new idea does prove robust, the team wishes to develop it further to make it even more effective against anticipated attacks, as well as other kinds of attacks they have not yet tested it against. At present, the defense only works on this specific kind of attack.

“Future attackers might try to consider attacks that can escape the feature-space noise we considered in this research,” said Ukita. “Indeed, attack and defense are two sides of the same coin; it’s an arms race that neither side will back down from, so we need to continually iterate, improve and innovate new ideas in order to protect the systems we use every day.”

The researchers began the work in the energy-producing portion of a cell, the mitochondria. In plants, aberrations within the mitochondrial genome can be associated with the inability to make pollen, a trait known as cytoplasmic male sterility (CMS) that has been successfully exploited for the production of high-yielding hybrid seeds in many important crops. Naturally occurring CMS-based systems that are robust enough to facilitate commercial scale hybrid seed production are limited, however.

In their proof-of-concept study, the NC State researchers, together with colleagues from Precision BioSciences and Elo Life Systems, deployed a unique strategy to test whether the CMS trait could be generated in tobacco, a commonly used model species in plant research. The researchers initially took an essential mitochondrial gene called atp1 and moved it to the nucleus after placing it under the regulatory control of an element — known as a promoter — that they predicted would allow the transferred atp1 gene to be expressed in every cell of the plant except those responsible for producing pollen. The researchers then used genome editing tools to permanently remove the native atp1 gene from the mitochondria.

Their approach was successful.

“The results exceeded our expectations,” said Ralph Dewey, Philip Morris Professor of Crop Science at NC State and corresponding author of a paper describing the research. “The plants looked completely normal up until they began to flower, but then failed to produce pollen because the transferred atp1 gene was no longer expressed. Importantly, because the original atp1 gene was deleted from the mitochondrial genome, the trait will be inherited maternally, which is a crucial consideration for large-scale hybrid seed production.”

Pollen wasn’t the only casualty of the technique. When cross-fertilized using pollen from a neighboring normal plant, their tobacco plants unexpectedly produced small, hollow seeds, much like those seen in popular “seedless” fruits such as watermelons and grapes.

“That’s because the promoter that we chose not only failed to express during pollen formation, but also during early seed development,” Dewey said.

Dewey said that his team is now working to decouple these results so that researchers can achieve either pollen infertility or the seedless trait alone, instead of both at the same time.

Dewey also stressed that the findings shouldn’t be limited to tobacco plants. Their next generation of experiments will include testing the seedless trait in tomato, a close relative of tobacco. They will also test their novel CMS trait in a grain such as rice to test the efficacy of their system in a crop where hybrid seed production is important for achieving maximal yields.

“Knowing the way the system works, there’s no reason to believe that we couldn’t effectively transfer the technology to other plant species,” he said.

The study appears in Frontiers in Plant Science. Devarshi Selote, H. Carol Griffin, Allison N. Dickey, Derek Jantz, J. Jeff Smith, Anna Matthiadis, Josh Strable, Caitlin Kestell and William A. Smith co-authored the paper. The work was funded in part by a grant from Elo Life Systems and supported by the NC State Plant Breeding Consortium. Dewey and several other researchers have filed a patent on the new technique.

Degeneration of dopaminergic neurons is widely accepted as the first event that leads to Parkinson’s. But the new study suggests that a dysfunction in the neuron’s synapses — the tiny gap across which a neuron can send an impulse to another neuron — leads to deficits in dopamine and precedes the neurodegeneration.

Parkinson’s disease affects 1% to 2% of the population and is characterized by resting tremor, rigidity and bradykinesia (slowness of movement). These motor symptoms are due to the progressive loss of dopaminergic neurons in the midbrain.

The findings, which will be published Sept. 15 in Neuron, open a new avenue for therapies, the scientists said.

“We showed that dopaminergic synapses become dysfunctional before neuronal death occurs,” said lead author Dr. Dimitri Krainc, chair of neurology at Northwestern University Feinberg School of Medicine and director of the Simpson Querrey Center for Neurogenetics. “Based on these findings, we hypothesize that targeting dysfunctional synapses before the neurons are degenerated may represent a better therapeutic strategy.”

The study investigated patient-derived midbrain neurons, which is critical because mouse and human dopamine neurons have a different physiology and findings in the mouse neurons are not translatable to humans, as highlighted in Krainc’s research recently published in Science.

Northwestern scientists found that dopaminergic synapses are not functioning correctly in various genetic forms of Parkinson’s disease. This work, together with other recent studies by Krainc’s lab, addresses one of the major gaps in the field: how different genes linked to Parkinson’s lead to degeneration of human dopaminergic neurons.

Neuronal recycling plant

Imagine two workers in a neuronal recycling plant. It’s their job to recycle mitochondria, the energy producers of the cell, that are too old or overworked. If the dysfunctional mitochondria remain in the cell, they can cause cellular dysfunction. The process of recycling or removing these old mitochondria is called mitophagy. The two workers in this recycling process are the genes Parkin and PINK1. In a normal situation, PINK1 activates Parkin to move the old mitochondria into the path to be recycled or disposed of.

It has been well-established that people who carry mutations in both copies of either PINK1 or Parkin develop Parkinson’s disease because of ineffective mitophagy.

The story of two sisters whose disease helped advance Parkinson’s research

Two sisters had the misfortune of being born without the PINK1 gene, because their parents were each missing a copy of the critical gene. This put the sisters at high risk for Parkinson’s disease, but one sister was diagnosed at age 16, while the other was not diagnosed until she was 48.

The reason for the disparity led to an important new discovery by Krainc and his group. The sister who was diagnosed at 16 also had partial loss of Parkin, which, by itself, should not cause Parkinson’s.

“There must be a complete loss of Parkin to cause Parkinson’s disease. So, why did the sister with only a partial loss of Parkin get the disease more than 30 years earlier?” Krainc asked.

As a result, the scientists realized that Parkin has another important job that had previously been unknown. The gene also functions in a different pathway in the synaptic terminal — unrelated to its recycling work — where it controls dopamine release. With this new understanding of what went wrong for the sister, Northwestern scientists saw a new opportunity to boost Parkin and the potential to prevent the degeneration of dopamine neurons.

“We discovered a new mechanism to activate Parkin in patient neurons,” Krainc said. “Now, we need to develop drugs that stimulate this pathway, correct synaptic dysfunction and hopefully prevent neuronal degeneration in Parkinson’s.”

The first author of the study is Pingping Song, research assistant professor in Krainc’s lab. Other authors are Wesley Peng, Zhong Xie, Daniel Ysselstein, Talia Krainc, Yvette Wong, Niccolò Mencacci, Jeffrey Savas, and D. James Surmeier from Northwestern and Kalle Gehring from McGill University.

The title of the article is “Parkinson’s disease linked parkin mutation disrupts recycling of synaptic vesicles in human dopaminergic neurons.”

This work was supported by National Institutes of Health grants R01NS076054, R3710 NS096241, R35 NS122257 and NS121174, all from the National Institute of Neurological Disorders and Stroke.

Jeyhan Kartaltepe, associate professor in the School of Physics and Astronomy, and Rebecca Larson, postdoctoral research associate, co-authored a paper, “Confirmation and refutation of very luminous galaxies in the early Universe,” published in Nature confirming very bright galaxies in the early universe, while also disproving the identification of what would have been the most distant galaxy ever found.

Kartaltepe and Larson, along with co-authors from around the world, studied the redshift (or displacement of the spectrum of an object toward longer, red wavelengths) of several specific galaxies to see how much the light shifted, which indicates how far away the galaxies are. The CEERS team focused on Maisie’s Galaxy, which was theorized to have a redshift of z ≈ 11.5, while a team in Scotland researched a nearby galaxy that they believed could have a redshift of z ≈ 16, far larger than any ever found before.

To examine further, the two teams partnered on a proposal to receive follow-up spectroscopy. When the new data came in, the teams were able to precisely measure the redshifts of both of these candidates, along with a few others.

“Spectra are how you really confirm what a galaxy’s redshift is,” explained Kartaltepe. “For these two galaxies, the answer was very clear — the spectra look completely different. We confirmed that Maisie’s Galaxy is at the high redshift we thought it was.”

The group also found that because of a coincidence that mimicked the colors of a high redshift galaxy, the other galaxy is not at a redshift of z ≈ 16, but at a redshift of z ≈ 4.9. Both the initial and follow-up data from JWST turned the theories into discoveries.

“Not only did JWST find these galaxies we didn’t know about before, but then it confirmed the redshift for them,” said Larson. “This paper in particular speaks to the power of not only JWST finding galaxies in the really early universe but also confirming and characterizing them.”

The research and the paper would not have been possible without dedicated collaboration between the CEERS team and the team in Scotland. Instead of working separately on their individual galaxies and submitting separate proposals, the partnership allowed for the follow-up spectroscopy to be accepted, and the subsequent analysis to be conducted efficiently, leading to new information about the universe.

When researching the data from JWST, scientists aim to find the highest redshift galaxies, or the most distant galaxies. Finding galaxies in the very early universe was one of the goals for the JWST. These and other early discoveries have proven the success of the telescope, even this early in its existence.

“It shows us how powerful the telescope is and its ability to do the things it was built to do,” said Kartaltepe. “In some ways, it performs even better than we expected. We are already learning so much about the universe early on in JWST’s mission. I think going forward we’re going to be able to build large samples over larger areas and really be able to dig deeper into the physical characteristics of galaxies in the early universe.”

The pregnancy complication affects up to eight per cent of pregnancies globally and is the leading cause of maternal and fetal mortality due to premature delivery, complications with the placenta and lack of oxygen.

The research, led by Drs. Kun Ping Lu and Xiao Zhen Zhou at Western, and Drs. Surendra Sharma and Sukanta Jash at Brown, has identified a toxic protein, cis P-tau, in the blood and placenta of preeclampsia patients.

According to the study published in Nature Communications, cis P-tau is a central circulating driver of preeclampsia — a “troublemaker” that plays a major role in causing the deadly complication.

“The root cause of preeclampsia has (so far) remained unknown, and without a known cause there has been no cure. Preterm delivery is the only life-saving measure,” said Lu, professor of biochemistry and oncology at Schulich School of Medicine & Dentistry. Lu is also a Western Research Chair in Biotherapeutics.

“Our study identifies cis P-tau as a crucial culprit and biomarker for preeclampsia. It can be used for early diagnosis of the complication and is a crucial therapeutic target,” said Sharma, who recently retired from his Brown roles as a professor of pathology and laboratory medicine (research) and professor of pediatrics (research).

In 2016, Sharma, a leading preeclampsia researcher, and his team had identified that preeclampsia and diseases like Alzheimer’s had similar root causes related to protein issues. This research builds on that finding.

Until now, cis P-tau was mainly associated with neurological disorders like Alzheimer’s disease, traumatic brain injuries (TBI) and stroke. This association was discovered by Lu and Zhou in 2015 as a result of their decades of research on the role of tau protein in cancer and Alzheimer’s.

An antibody developed by Zhou in 2012 to target only the toxic protein while leaving its healthy counterpart unscathed is currently undergoing clinical trials in human patients suffering from TBI and Alzheimer’s Disease. The antibody has shown promising results in animal models and human cell cultures in treating the brain conditions.

The researchers were curious whether the same antibody could work as a potential treatment for preeclampsia. Upon testing the antibody in mouse models they found astonishing results.

“In this study, we found the cis P-tau antibody efficiently depleted the toxic protein in the blood and placenta, and corrected all features associated with preeclampsia in mice. Clinical features of preeclampsia, like elevated blood pressure, excessive protein in urine and fetal growth restriction, among others, were eliminated and pregnancy was normal,” said Sharma.

Sharma and his team at Brown have been working on developing an assay for early detection of preeclampsia and therapies to treat the condition. He believes the findings of this study have brought them closer to their goal.

Black and Hispanic women more susceptible

The tragic death of American track and field champion Tori Bowie earlier this year put the spotlight on preeclampsia, which disproportionately impacts Black and Hispanic women.

A gold, silver and bronze medalist in the 2016 Olympic Games, Bowie, 32, was found dead in her bed on May 2, 2023, while approximately eight months pregnant. According to the autopsy report the complications may have involved eclampsia — a severe form of preeclampsia.

“Research has shown that women of certain races have genes that could possibly lead to higher than average blood pressure levels, eventually creating conditions for preeclampsia during pregnancy. However, it’s also true that in many low socio-economic countries there’s no registry to record PE cases. So, its link to other environmental factors is still unclear,” said Sharma.

Preeclampsia and the brain

Recent research has also thrown light on preeclampsia’s long-term impacts and possible links to brain health.

“Preeclampsia presents immediate dangers to both the mother and fetus, but its long-term effects are less understood and still unfolding,” said Sharma. “Research has suggested a heightened risk of dementia later in life for both mothers who have experienced preeclampsia and their children.” However, the causal link between preeclampsia and dementia is not known.

The researchers say this new study has pinpointed a potential underlying cause of the complex relationship between preeclampsia and brain health.

“Our study adds another layer to this complexity. For the first time, we’ve identified significant levels of cis P-tau outside the brain in the placenta and blood of preeclampsia patients. This suggests a deeper connection between preeclampsia and brain-related issues,” said Jash, the lead author of the study.

As researchers delve deeper, how our bodies respond to stress is also emerging as a potential factor in the onset of preeclampsia.

“Although genetics play a role, factors like stress could be an important piece of the puzzle. Understanding how stress and other environmental factors intersect with biological markers like cis P-tau may offer a more complete picture,” said Jash, assistant professor of molecular biology, cell biology and biochemistry (research) and pediatrics (research) at Brown.

A stress-response enzyme called Pin1

In 1996 and 1997, Lu and Zhou made the groundbreaking discovery of Pin1, which turns out to be a stress-response enzyme. This is a specific protein in the cells that becomes active or changes its behaviour in response to stressors, such as environmental challenges, toxins or physiological changes.

“Pin1 plays a pivotal role in keeping proteins, including the tau protein, in the functional shape during stress. When Pin1 becomes inactivated, it leads to the formation of a toxic, misshapen, variant of tau — cis P-tau,” said Zhou, associate professor, pathology and laboratory medicine at Schulich Medicine & Dentistry.

Interestingly, Pin1 is a key player in cancer signalling networks, turning on numerous cancer-causing proteins and turning off many cancer-suppressing ones. Found in high levels in most human cancers, it’s particularly active in cancer stem cells, which are thought to be central to starting and spreading tumors and are hard to target with existing treatments.

“Essentially, when Pin1 is activated, it can lead to cancer. On the other hand, when there’s a decrease or deactivation in Pin1, it results in the formation of the toxic protein cis P-tau, which leads to memory loss in Alzheimer’s and after TBI or stroke. Now, we’ve uncovered its connection to preeclampsia as well,” said Zhou.

“The results have far-reaching implications. This could revolutionize how we understand and treat a range of conditions, from pregnancy-related issues to brain disorders,” said Lu.

Lu and Sharma had met at Brown in 2019, where Lu was invited to give a lecture on his research. Following an engaging session and a few dinners together, a collaboration between the Western researchers and Brown was forged.

“Science surprises us. I had never thought of working on finding a therapy for preeclampsia. It also shows that a collaboration can be transformative.”

Led by Saptarshi Das, associate professor of engineering science and mechanics at Penn State, the team published their work on September 15 in Nature Communication.

“Robots make decisions based on the environment they are in, but their sensors do not generally talk to each other,” said Das, who also has joint appointments in electrical engineering and in materials science and engineering. “A collective decision can be made through a sensor processing unit, but is that the most efficient or effective method? In the human brain, one sense can influence another and allow the person to better judge a situation.”

For instance, a car might have one sensor scanning for obstacles, while another senses darkness to modulate the intensity of the headlights. Individually, these sensors relay information to a central unit which then instructs the car to brake or adjust the headlights. According to Das, this process consumes more energy. Allowing sensors to communicate directly with each other can be more efficient in terms of energy and speed — particularly when the inputs from both are faint.

“Biology enables small organisms to thrive in environments with limited resources, minimizing energy consumption in the process,” said Das, who is also affiliated with the Materials Research Institute. “The requirements for different sensors are based on the context — in a dark forest, you’d rely more on listening than seeing, but we don’t make decisions based on just one sense. We have a complete sense of our surroundings, and our decision making is based on the integration of what we’re seeing, hearing, touching, smelling, etcetera. The senses evolved together in biology, but separately in AI. In this work, we’re looking to combine sensors and mimic how our brains actually work.”

The team focused on integrating a tactile sensor and a visual sensor so that the output of one sensor modifies the other, with the help of visual memory. According to Muhtasim Ul Karim Sadaf, a third-year doctoral student in engineering science and mechanics, even a short-lived flash of light can significantly enhance the chance of successful movement through a dark room.

“This is because visual memory can subsequently influence and aid the tactile responses for navigation,” Sadaf said. “This would not be possible if our visual and tactile cortex were to respond to their respective unimodal cues alone. We have a photo memory effect, where light shines and we can remember. We incorporated that ability into a device through a transistor that provides the same response.”

The researchers fabricated the multisensory neuron by connecting a tactile sensor to a phototransistor based on a monolayer of molybdenum disulfide, a compound that exhibits unique electrical and optical characteristics useful for detecting light and supporting transistors. The sensor generates electrical spikes in a manner reminiscent of neurons processing information, allowing it to integrate both visual and tactile cues.

It’s the equivalent of seeing an “on” light on the stove and feeling heat coming off of a burner — seeing the light on doesn’t necessarily mean the burner is hot yet, but a hand only needs to feel a nanosecond of heat before the body reacts and pulls the hand away from the potential danger. The input of light and heat triggered signals that induced the hand’s response. In this case, the researchers measured the artificial neuron’s version of this by seeing signaling outputs resulted from visual and tactile input cues.

To simulate touch input, the tactile sensor used triboelectric effect, in which two layers slide against one another to produce electricity, meaning the touch stimuli was encoded into electrical impulses. To simulate visual input, the researchers shined a light into the monolayer molybdenum disulfide photo memtransistor — or a transistor that can remember visual input, like how a person can hold onto the general layout of a room after a quick flash illuminates it.

They found that the sensory response of the neuron — simulated as electrical output — increased when both visual and tactile signals were weak.

“Interestingly, this effect resonates remarkably well with its biological counterpart — a visual memory naturally enhances the sensitivity to tactile stimulus,” said co-first author Najam U Sakib, a third-year doctoral student in engineering science and mechanics. “When cues are weak, you need to combine them to better understand the information, and that’s what we saw in the results.”

Das explained that an artificial multisensory neuron system could enhance sensor technology’s efficiency, paving the way for more eco-friendly AI uses. As a result, robots, drones and self-driving vehicles could navigate their environment more effectively while using less energy.

“The super additive summation of weak visual and tactile cues is the key accomplishment of our research,” said co-author Andrew Pannone, a fourth-year doctoral student in engineering science and mechanics. “For this work, we only looked into two senses. We’re working to identify the proper scenario to incorporate more senses and see what benefits they may offer.”

Harikrishnan Ravichandran, a fourth-year doctoral student in engineering science and mechanics at Penn State, also co-authored this paper.

The Army Research Office and the National Science Foundation supported this work.