“This is a tale of a failed galaxy,” said the program’s principal investigator, Alejandro Benitez-Llambay of the Milano-Bicocca University in Milan, Italy. “In science, we usually learn more from the failures than from the successes. In this case, seeing no stars is what proves the theory right. It tells us that we have found in the local Universe a primordial building block of a galaxy that hasn’t formed.”

A Rare Glimpse of the Dark Universe

“This cloud is a window into the dark Universe,” explained team member Andrew Fox of AURA/STScI for the European Space Agency. “We know from theory that most of the mass in the Universe is expected to be dark matter, but it’s difficult to detect this dark material because it doesn’t emit light. Cloud-9 gives us a rare look at a dark-matter-dominated cloud.”

Cloud-9 belongs to a category known as Reionization-Limited H I Clouds, or “RELHICs.” The term “H I” refers to neutral hydrogen, while “RELHIC” describes a hydrogen cloud formed in the early Universe that never progressed to form stars. Scientists had predicted the existence of such objects for years, but direct confirmation remained elusive. Only after observing Cloud-9 with Hubble were researchers able to verify that it truly contains no stars.

Ruling Out a Hidden Dwarf Galaxy

“Before we used Hubble, you could argue that this is a faint dwarf galaxy that we could not see with ground-based telescopes. They just didn’t go deep enough in sensitivity to uncover stars,” explained lead author Gagandeep Anand of the Space Telescope Science Institute (STScI), Baltimore, USA. “But with Hubble’s Advanced Camera for Surveys, we’re able to nail down that there’s nothing there.”

The discovery itself caught researchers off guard. “Among our galactic neighbors, there might be a few abandoned houses out there,” said STScI’s Rachael Beaton, who is also on the research team.

RELHICs are believed to be dark matter structures that never gathered enough gas to trigger star formation. Because of this, they preserve conditions from the early Universe. Cloud-9 points to the likely existence of many other small, dark matter-dominated objects, often described as failed galaxies. Studying them offers new insight into parts of the Universe that remain difficult to observe because most telescopes focus on bright stars and galaxies.

Measuring an Invisible Giant

Hydrogen clouds near the Milky Way have been studied for decades, but most are far larger and more irregular in shape than Cloud-9. By contrast, Cloud-9 is smaller, denser, and nearly spherical, giving it a distinctly different appearance from other known gas clouds.

At its center, Cloud-9 contains neutral hydrogen spanning roughly 4900 light-years. The hydrogen gas alone has a mass about 1 million times that of the Sun. If the gas pressure is indeed balanced by the gravitational pull of the surrounding dark matter, then dark matter must account for most of the object’s mass. Based on this balance, Cloud-9 is estimated to contain roughly 5 billion solar masses.

Why Starless Objects Matter

Cloud-9 highlights how much of the Universe exists beyond stars. Observing starlight alone does not reveal the full picture. By examining gas and dark matter together, scientists can better understand systems that would otherwise remain hidden.

Finding failed galaxies like Cloud-9 is difficult because nearby bright objects often overpower their faint signals. These systems are also sensitive to environmental effects such as ram-pressure stripping, which can remove gas as they move through intergalactic space. These challenges help explain why such objects appear to be rare.

Discovery Through Radio Telescopes

Cloud-9 was first detected three years ago during a radio survey conducted with the Five-hundred-meter Aperture Spherical Telescope (FAST) in Guizhou, China. The discovery was later confirmed using the Green Bank Telescope and the Very Large Array in the United States. The name “Cloud-9” carries no cultural meaning in China and was assigned simply because it was the ninth gas cloud identified near the outer regions of the spiral galaxy Messier 94 (M94).

The cloud lies close to M94 and appears to be physically connected to the galaxy. High-resolution radio observations reveal slight distortions in the gas, which may be evidence of interaction between Cloud-9 and its larger neighbor.

A Galaxy That Might Still Form

Whether Cloud-9 will eventually become a galaxy depends on whether it gains additional mass. If it had been much larger, gravity would have caused it to collapse and form stars long ago. If it were much smaller, its gas might have dispersed and become ionized, leaving little behind. Instead, it exists in a narrow range that allows it to persist as a RELHIC.

This discovery advances understanding of how galaxies form, how the early Universe evolved, and how dark matter behaves. Because Cloud-9 contains no stars, it allows scientists to study the properties of dark matter clouds without interference from starlight. As future surveys improve, researchers expect to uncover more of these rare relics, offering deeper insight into the Universe’s hidden structure and the physics of dark matter.

Expanded DNA repeats are responsible for more than 60 inherited disorders. These conditions develop when repeating genetic sequences lengthen beyond normal limits and interfere with healthy cell function. Examples include Huntington’s disease, myotonic dystrophy, and certain forms of ALS.

Although most people carry DNA repeats that slowly expand throughout life, scientists had not previously examined how widespread this instability is or which genes control it using large biobank datasets. This research shows that repeat expansion is far more common than previously recognized. It also identifies dozens of genes involved in regulating the process, creating new opportunities to develop treatments that could slow disease progression.

How Researchers Studied Nearly a Million Genomes

The research team, which included scientists from UCLA, the Broad Institute, and Harvard Medical School, analyzed whole genome sequencing data from 490,416 participants in the UK Biobank and 414,830 participants in the All of Us Research Program. To carry out the analysis, they developed new computational approaches capable of measuring DNA repeat length and instability using standard sequencing data.

Using these tools, the team examined 356,131 variable repeat sites across the human genome. They tracked how repeat lengths changed with age in blood cells and identified inherited genetic variants that affected the speed of expansion. The researchers also searched for associations between repeat expansion and thousands of disease outcomes in order to uncover previously unknown links to human illness.

Key Findings on DNA Repeat Instability

The study found that common DNA repeats in blood cells consistently expand as people get older. Researchers identified 29 regions of the genome where inherited genetic variants altered repeat expansion rates, with differences of up to fourfold between individuals with the highest and lowest genetic risk scores.

One surprising result was that the same DNA repair genes did not behave uniformly. Genetic variants that helped stabilize some repeats made other repeats more unstable. The researchers also identified a newly recognized repeat expansion disorder involving the GLS gene. Expansions in this gene, which occur in about 0.03% of people, were linked to a 14-fold increase in the risk of severe kidney disease and a 3-fold increase in the risk of liver diseases.

What the Findings Mean for Future Research

The results suggest that measuring DNA repeat expansion in blood could serve as a useful biomarker for evaluating future treatments designed to slow repeat growth in diseases such as Huntington’s. The computational tools developed for this study can now be applied to other large biobank datasets to identify additional unstable DNA repeats and related disease risks.

Researchers note that further mechanistic studies will be needed to understand why the same genetic modifiers can have opposite effects on different repeats. These efforts will focus on how DNA repair processes differ across cell types and genetic contexts. The discovery of kidney and liver disease linked to GLS repeat expansion also suggests that additional, previously unrecognized repeat expansion disorders may be hidden within existing genetic data.

Expert Perspective on the Findings

“We found that most human genomes contain repeat elements that expand as we age,” said Margaux L. A. Hujoel, PhD, lead author of the study and assistant professor in the Departments of Human Genetics and Computational Medicine at the David Geffen School of Medicine at UCLA. “The strong genetic control of this expansion, with some individuals’ repeats expanding four times faster than others, points to opportunities for therapeutic intervention. These naturally occurring genetic modifiers show us which molecular pathways could be targeted to slow repeat expansion in disease.”

Margaux L. A. Hujoel (UCLA and Brigham and Women’s Hospital/Harvard Medical School), Robert E. Handsaker (Broad Institute and Harvard Medical School), David Tang (Brigham and Women’s Hospital/Harvard Medical School), Nolan Kamitaki (Brigham and Women’s Hospital/Harvard Medical School), Ronen E. Mukamel (Brigham and Women’s Hospital/Harvard Medical School), Simone Rubinacci (Brigham and Women’s Hospital/Harvard Medical School and Institute for Molecular Medicine Finland), Pier Francesco Palamara (University of Oxford), Steven A. McCarroll (Broad Institute and Harvard Medical School), Po-Ru Loh (Brigham and Women’s Hospital/Harvard Medical School and Broad Institute)

M.L.A.H. was supported by US NIH fellowship F32 HL160061; R.E.H. and S.A.M. by US NIH grant R01 HG006855; D.T. by US NIH training grant T32 HG002295; N.K. by US NIH training grant T32 HG002295 and fellowship F31 DE034283; R.E.M. by US NIH grant K25 HL150334; S.R. by a Swiss National Science Foundation Postdoc. Mobility fellowship; P.F.P. by ERC Starting Grant no. 850869; and P.-R.L. by US NIH grants R56 HG012698, R01 HG013110 and UM1 DA058230 and a Burroughs Wellcome Fund Career Award. The All of Us Research Program is supported by the NIH. The authors declare no competing interests.

Preservatives are part of the broader category of food additives and are widely used throughout the global food supply. Their presence is extensive. In 2024, the Open Food Facts World database listed around three and a half million food and beverage products. More than 700,000 of those products contained at least one preservative.

Two Major Types of Preservative Additives

In their analysis, Inserm researchers divided preservative additives into two main groups. The first group includes non-antioxidant preservatives, which slow spoilage by limiting microbial growth or slowing chemical reactions in food. The second group consists of antioxidant additives, which help preserve foods by reducing or controlling exposure to oxygen in packaging.

On ingredient labels, these additives typically appear under European codes between E200 and E299 (for preservatives in the strict sense) and between E300 and E399 (for antioxidant additives).

Why Researchers Are Investigating Preservatives

Earlier experimental research has raised concerns that some preservatives may harm cells or DNA and interfere with normal metabolic processes. However, direct evidence linking preservative intake to type 2 diabetes in large human populations has been limited until now.

To better understand this potential connection, a research team led by Mathilde Touvier, Inserm Research Director, examined long-term exposure to food preservatives and the incidence of type 2 diabetes using detailed data from the NutriNet-Santé study.

Tracking Diet and Health Over More Than a Decade

The study followed more than 100,000 French adults between 2009 and 2023. Participants regularly provided information about their medical history, socio-demographic background, physical activity, lifestyle habits, and overall health.

They also submitted detailed food records covering multiple 24-hour periods. These records included the names and brands of industrial food products they consumed. Researchers cross-referenced this information with several databases (Open Food Facts, Oqali, EFSA) and combined it with measurements of additives in foods and beverages. This allowed the team to estimate each participant’s long-term exposure to preservatives.

Measuring Preservative Consumption

Across all food records, researchers identified a total of 58 preservative-related additives. This included 33 preservatives in the strict sense and 27 antioxidant additives. From this group, 17 preservatives were analyzed individually because they were consumed by at least 10% of the study participants.

The analysis accounted for many factors that could influence diabetes risk, including age, sex, education, smoking, alcohol use, and overall diet quality (calories, sugar, salt, saturated fats, fibre, etc.).

Diabetes Cases and Risk Increases

Over the study period, 1,131 cases of type 2 diabetes were identified among the 108,723 participants.

Compared with people who consumed the lowest levels of preservatives, those with higher intake showed a markedly increased risk of developing type 2 diabetes. Overall preservative consumption was linked to a 47% higher risk. Non-antioxidant preservatives were associated with a 49% increase, while antioxidant additives were linked to a 40% higher risk.

Specific Preservatives Associated With Risk

Among the 17 preservatives examined individually, higher intake of 12 was associated with an increased risk of type 2 diabetes. These included widely used non-antioxidant preservatives (potassium sorbate (E202), potassium metabisulphite (E224), sodium nitrite (E250), acetic acid (E260), sodium acetates (E262) and calcium propionate (E282)) as well as antioxidant additives (sodium ascorbate (E301), alpha-tocopherol (E307), sodium erythorbate (E316), citric acid (E330), phosphoric acid (E338) and rosemary extracts (E392)).

What the Researchers Say

“This is the first study in the world on the links between preservative additives and the incidence of type 2 diabetes. Although the results need to be confirmed, they are consistent with experimental data suggesting the harmful effects of several of these compounds,” explains Mathilde Touvier, Inserm research director and coordinator of this work.

“More broadly, these new data add to others in favor of a reassessment of the regulations governing the general use of food additives by the food industry in order to improve consumer protection,” adds Anaïs Hasenböhler, a doctoral student at EREN who conducted these studies.

“This work once again justifies the recommendations made by the National Nutrition and Health Programme to consumers to favor fresh, minimally processed foods and to limit unnecessary additives as much as possible,” concludes Mathilde Touvier.

This work was funded by the European Research Council (ERC ADDITIVES), the National Cancer Institute, and the French Ministry of Health.

The system, called SleepFM, was trained using almost 600,000 hours of sleep recordings from 65,000 individuals. These recordings came from polysomnography, an in-depth sleep test that uses multiple sensors to track brain activity, heart function, breathing patterns, eye movement, leg motion, and other physical signals during sleep.

Sleep Studies Hold Untapped Health Data

Polysomnography is considered the gold standard for evaluating sleep and is typically performed overnight in a laboratory setting. While it is widely used to diagnose sleep disorders, researchers realized it also captures a vast amount of physiological information that has rarely been fully analyzed.

“We record an amazing number of signals when we study sleep,” said Emmanual Mignot, MD, PhD, the Craig Reynolds Professor in Sleep Medicine and co-senior author of the new study, which will publish Jan. 6 in Nature Medicine. “It’s a kind of general physiology that we study for eight hours in a subject who’s completely captive. It’s very data rich.”

In routine clinical practice, only a small portion of this information is examined. Recent advances in artificial intelligence now allow researchers to analyze these large and complex datasets more thoroughly. According to the team, this work is the first to apply AI to sleep data on such a massive scale.

“From an AI perspective, sleep is relatively understudied. There’s a lot of other AI work that’s looking at pathology or cardiology, but relatively little looking at sleep, despite sleep being such an important part of life,” said James Zou, PhD, associate professor of biomedical data science and co-senior author of the study.

Teaching AI the Patterns of Sleep

To unlock insights from the data, the researchers built a foundation model, a type of AI designed to learn broad patterns from very large datasets and then apply that knowledge to many tasks. Large language models like ChatGPT use a similar approach, though they are trained on text rather than biological signals.

SleepFM was trained on 585,000 hours of polysomnography data collected from patients evaluated at sleep clinics. Each sleep recording was divided into five-second segments, which function much like words used to train language-based AI systems.

“SleepFM is essentially learning the language of sleep,” Zou said.

The model integrates multiple streams of information, including brain signals, heart rhythms, muscle activity, pulse measurements, and airflow during breathing, and learns how these signals interact. To help the system understand these relationships, the researchers developed a training method called leave-one-out contrastive learning. This approach removes one type of signal at a time and asks the model to reconstruct it using the remaining data.

“One of the technical advances that we made in this work is to figure out how to harmonize all these different data modalities so they can come together to learn the same language,” Zou said.

Predicting Future Disease From Sleep

After training, the researchers adapted the model for specific tasks. They first tested it on standard sleep assessments, such as identifying sleep stages and evaluating sleep apnea severity. In these tests, SleepFM matched or exceeded the performance of leading models currently in use.

The team then pursued a more ambitious objective: determining whether sleep data could predict future disease. To do this, they linked polysomnography records with long-term health outcomes from the same individuals. This was possible because the researchers had access to decades of medical records from a single sleep clinic.

The Stanford Sleep Medicine Center was founded in 1970 by the late William Dement, MD, PhD, who is widely regarded as the father of sleep medicine. The largest group used to train SleepFM included about 35,000 patients between the ages of 2 and 96. Their sleep studies were recorded at the clinic between 1999 and 2024 and paired with electronic health records that followed some patients for as long as 25 years.

(The clinic’s polysomnography recordings go back even further, but only on paper, said Mignot, who directed the sleep center from 2010 to 2019.)

Using this combined dataset, SleepFM reviewed more than 1,000 disease categories and identified 130 conditions that could be predicted with reasonable accuracy using sleep data alone. The strongest results were seen for cancers, pregnancy complications, circulatory diseases, and mental health disorders, with prediction scores above a C-index of 0.8.

How Prediction Accuracy Is Measured

The C-index, or concordance index, measures how well a model can rank people by risk. It reflects how often the model correctly predicts which of two individuals will experience a health event first.

“For all possible pairs of individuals, the model gives a ranking of who’s more likely to experience an event — a heart attack, for instance — earlier. A C-index of 0.8 means that 80% of the time, the model’s prediction is concordant with what actually happened,” Zou said.

SleepFM performed especially well when predicting Parkinson’s disease (C-index 0.89), dementia (0.85), hypertensive heart disease (0.84), heart attack (0.81), prostate cancer (0.89), breast cancer (0.87), and death (0.84).

“We were pleasantly surprised that for a pretty diverse set of conditions, the model is able to make informative predictions,” Zou said.

Zou also noted that models with lower accuracy, often around a C-index of 0.7, are already used in medical practice, such as tools that help predict how patients might respond to certain cancer treatments.

Understanding What the AI Sees

The researchers are now working to improve SleepFM’s predictions and better understand how the system reaches its conclusions. Future versions may incorporate data from wearable devices to expand the range of physiological signals.

“It doesn’t explain that to us in English,” Zou said. “But we have developed different interpretation techniques to figure out what the model is looking at when it’s making a specific disease prediction.”

The team found that while heart-related signals were more influential in predicting cardiovascular disease and brain-related signals played a larger role in mental health predictions, the most accurate results came from combining all types of data.

“The most information we got for predicting disease was by contrasting the different channels,” Mignot said. Body constituents that were out of sync — a brain that looks asleep but a heart that looks awake, for example — seemed to spell trouble.

Rahul Thapa, a PhD student in biomedical data science, and Magnus Ruud Kjaer, a PhD student at Technical University of Denmark, are co-lead authors of the study.

Researchers from the Technical University of Denmark, Copenhagen University Hospital -Rigshospitalet, BioSerenity, University of Copenhagen and Harvard Medical School contributed to the work.

The study received funding from the National Institutes of Health (grant R01HL161253), Knight-Hennessy Scholars and Chan-Zuckerberg Biohub.