EU tightens rules on 'forever chemicals' in drinking water

The European Union on Monday began enforcing stricter limits on PFAS “forever chemicals” in drinking water, requiring authorities across all member states to test tap water for contamination.


Issued on: 12/01/2026 - RFI

Tap water across the European Union is now subject to stricter testing rules for PFAS, a group of highly persistent chemicals known as “forever chemicals”. AP - Attila Kovacs

The measures stem from the recast EU Drinking Water Directive adopted in 2020. They set two safety limits for PFAS in drinking water: one covering 20 of the most concerning chemicals, and another covering all PFAS combined.

The rules require authorities to act if those limits are exceeded.

PFAS, or per- and polyfluoroalkyl substances, are a large group of human-made chemicals used in products such as pesticides, waterproof coatings and industrial materials. They break down very slowly and can remain in the environment for many years.

In France, the new EU rules follow a large national testing campaign carried out by Anses, France’s food and health safety agency, between 2023 and 2025. The aim was to get a clearer picture of PFAS contamination in drinking water before the EU deadline.

France cracks down on 'forever chemicals' in cosmetics, clothing

Nationwide snapshot

Anses screened for 35 PFAS substances in more than 600 samples of raw water and the same number of tap water samples. The testing covered mainland France and overseas territories.

Two-thirds of the samples were taken from areas where regional health agencies had identified a potential risk of PFAS pollution.

Of the 35 PFAS screened for, 20 were detected at least once in raw water and 19 in tap water.

Anses said that in the vast majority of samples, PFAS levels were below the regulatory limits that apply. Only a small number exceeded the EU threshold for the combined level of the 20 PFAS covered by the directive.

The agency said the results would help improve long-term monitoring and decide which substances may need closer attention in future.

For environmental group Générations Futures, the new rules mark a major change in how contamination is tracked nationwide.

“Finally, we are going to have data on the presence of the 20 main PFAS in tap water for the whole of France,” François Veillerette, spokesperson for Générations Futures, told France Culture radio.

Until now, monitoring had been uneven.

“In 2025, we had data in some regions, but not everywhere,” Veillerette said, adding that the new system would allow comparisons between regions, tracking over time and identification of areas most at risk.

French tap water tainted by widespread forever chemicals, study finds

A chemical left out

One substance stood out in the Anses campaign. Trifluoroacetic acid, known as TFA, was detected in 92 percent of both raw and mains-supplied water samples. Its concentration varied widely from place to place.

TFA is not included in the list of 20 PFAS covered by the new EU limit.

“It will be monitored next year, but we would have liked it to be included from 2026,” said Veillerette.

He warned that once PFAS pollution becomes established, treating drinking water becomes expensive. “If these substances settle in and increase, it will cost a lot to filter and treat the water,” he said.

Big Data Is a Bad Idea: Why AI Factory Farms Will Not Save Rural America

AI data centers have been added to the limited menu for economic development in marginalized US communities, but people in those communities have good reason to oppose them.


A sign on a rural Michigan road opposes a planned $7 billion data center on southeast Michigan farm land in Saline, Michigan on December 1, 2025.
(Photo by Jim West/UCG/Universal Images Group via Getty Images)

John Peck
Jan 12, 2026

Common Dreams


One word—plastics. That was the golden grail that Dustin Hoffman learned about from some well-wisher in the movie The Graduate. I remember watching the film as a farm kid and thinking about the updated version I was being told by my guidance counselors—one word: computers. We are now in the midst of the “Fourth Industrial Revolution” and the latest mantra is: artificial intelligence. Such free advice, though, could really be a costly warning in disguise.

Granted, there is a lot of poverty in the “richest” nation on Earth, and marginalized US communities often have few choices for economic (mal) development. It becomes a twisted game of pick your own poison: supermax prison, toxic waste dump, ethanol facility, tar sands pipeline… Now, AI data centers have been added to the limited menu. Someone recently shared a map of looming AI data centers across the world. It reminded me of how a tumor spreads and Edward Abbey’s quote that “growth for the sake of growth is the ideology of the cancer cell.”



Big Tech Ramps Up Propaganda Blitz As AI Data Centers Become Toxic With Voters



US Electric Grid Heading Toward ‘Crisis’ Thanks to AI Data Centers

The fact that Big Data has targeted Rural America for its latest mastitis should be no surprise. We have lots of available land to grab, thanks to the legacy of settler colonialism and family-farm foreclosure. Back in August I remember driving past Beaver Dam, Wisconsin and watching bulldozers flattening over 800 acres along Hwy 151 and my first hunch was: data center. Sure enough, the secretive $1 billion deal with Meta was finally revealed in a November press release. Just north of Madison in the town of DeForest, Blackstone subsidiary QTS Realty Trust is aiming to build another $12 billion data center on close to 1,600 acres. And if we need to free up more land for AI, we quaint rural folks could just abandon growing real Xmas trees and force people to buy plastic ones instead, as one Fox News “expert” suggested over the holidays. Former President Joe Biden visited Mt. Pleasant, Wisconsin in May 2024 to promote Microsoft’s new $3.3 billion 300+ acre AI campus on the former site of flat screen maker, Foxconn, that welcomed President Donald Trump for its groundbreaking back in 2018. Foxconn abandoned that $10 billion project and its 13,000 job promise, after getting millions in state subsidies and local tax deferrals.

The Microsoft AI complex in Mt. Pleasant will also require over 8 million gallons of water per year from Lake Michigan. We still have some clean water, though that may not last long thanks to agrochemical monocultures, CAFO manure dumping, and PFAS-laden sludge spreading. And AI certainly is thirsty—the Alliance for the Great Lakes noted in its August 2025 report that a hyperscale AI data center needs up to 365 million gallons of water to keep itself cool—that is as much water as is needed by 12,000 people! A recent investigative report by Bloomberg News found that over two-thirds of the AI data centers built since 2022 are in parts of the country already facing water stress. And it is really hard to drink data.

But is all the AI hype just another bubble about to burst? Rural communities (and public taxpayers) have been offered many “amazing” schemes in the past that ended up being just a “bait and switch”—another hollow promise.

In the Midwest we also have potential access to vast electricity (fracked natural gas, wind and solar farms, methane digesters), and relatively under-stressed high voltage grids (unlike California or Texas), though the loss of “cheaper” imported Canadian hydropower with the latest trade war could be a serious challenge. In 2023 the US had over a $2 billion electricity trade deficit vis-a-vis Canada. According to a recent Clean Wisconsin report, just two of our proposed AI data centers will require 3.9 gigawatts—1.5 times the current power demand of all 4.3 million homes in the state.

But, no worry, there are dilapidated US nuclear reactors with massive waste dumps that could be put back online such as Palisades in Michigan, despite opposition from environmental activists and family farmers. The Trump administration also just announced a $1 billion low-interest loan to reanimate Three Mile Island in Pennsylvania for the sake of AI. Until all that happens, though, regular ratepayers can expect a huge hike in their energy bills as Big Data has the market clout to siphon off what it needs first, especially as it colludes with utility monopolies. Many people in Wisconsin are already paying for $1+ billion in stranded assets—mostly defunct coal plants, as well as nuclear waste storage facilities—while utility investors continue to receive guaranteed dividends of 9-10%.

But is all the AI hype just another bubble about to burst? Rural communities (and public taxpayers) have been offered many “amazing” schemes in the past that ended up being just a “bait and switch”—another hollow promise. If we subsidize a massive data center, will the projected “market” for increasing algorithms actually come? Many within the AI industry don’t think so, and are now invoking the lessons we should have learned from the Enron scandal decades ago or the even worse sequel in the subprime mortgage-fueled financial meltdown. Corporate cheerleaders can be quite clever when it comes to inflating prices (and stocks) for goods and services that may not even exist, while hiding their massive debt obligations in a whole cascading series of shadowy shell subsidiaries and dishonest accounting shenanigans.

Many industry insiders are ringing alarm bells. “These models are being hyped up, and we’re investing more than we should,” said Daron Acemoglu, who won the 2024 Nobel Economics Prize, quoted in a recent NPR story about the current AI boom or bubble. OpenAI says it will spend $1.4 trillion on data centers over the next eight years, while Amazon, Google, Meta, and Microsoft are going to throw in another $400 billion. Meanwhile, just 3% of people who use AI now pay for it, and many are frantically trying to figure out how to turn off AI mode on their internet searches and to reject AI eavesdropping on their Zoom calls. Where is the real revenue going to come from to pay for all this AI speculation? The same NPR story notes that such a flood of leveraged capital is equal to every iPhone user on Earth forking over $250 to “enjoy” the benefits of AI—and “that’s not going to happen,” adds Paul Kedrosky, a venture capitalist who is now a research fellow at MIT’s Institute for the Digital Economy. Morgan Stanley estimates AI companies will shell out $3 trillion by 2028 for this data center buildout—but less than 50% of that money will come from them. Hmmm...

Special purpose vehicle (SPV) may sound like a fancy name for a retrofitted tractor, but that is how Big Data is creating a Potemkin Village to hide their Ponzi Scheme. Here is one example from Richland Parish, Louisiana where Meta is now building its Hyperion Data Center—a massive $27 billion project. A Wall Street outfit, Blue Owl, borrows $27 billion, using Meta’s future rent payments for a data center to back up its loan. Meta’s 20% “mortgage” on the facility gives them 100% control of the purported data crunching from the facility. This debt never shows up on Meta’s books and remains hidden from carefree investors and shallow analysts, but, like other synthetic financial instruments such as the now infamous mortgage backed security (MBS), the reality only comes home to roost when the house of cards collapses and Meta has to eventually pay off Blue Owl.

In the meantime, as the Louisiana Illuminator reports, the residents of Richland Parish (where 25% live below the poverty level) are bearing the brunt of all the real costs of having an AI factory farm. Dozens of crashes involving construction vehicles; damage to local roads; and massive future energy demands (three times that required for the entire city of New Orleans), which will entail new natural gas power plants to be built (subsidized by existing ratepayers even as fossil fuel-induced climate change floods the Louisiana delta). Beyond the initial building flurry, AI data centers are ultimately job poor. It just doesn’t take that many people to tend computers once they are built. As Meta’s VP, Brad Smith, admitted, the 250,000 square foot Hyperion data center may need 1,500 workers to build but barely 50 to operate. Beyond all the ballyhoo, the main reason a particular community is chosen to “host” one seems to be based upon the bought duplicity of elected officials and the excessive generosity of local taxpayers. Not a good cost-benefit analysis—unless you are Big Data.

And then there are the questionable kickback schemes between the suppliers of the technology and those owning the data centers. If you are maker of computer chips, would you not be tempted to fork over capital to a major buyer of your own products to ensure future demand? Nvidia just announced a $100 billion stake in OpenAI to help bankroll the data centers. In turn OpenAI signed a $300 billion deal with Oracle to actually build the AI data centers that will require Nvidia’s graphics processing units (GPUs). OpenAI also signed a separate $6+ billion deal with former BitCoin miner, CoreWeave, which rents out internet cloud access (using Nvidia’s chips once again). This type of incestuous circular financing should raise eyebrows to anyone who studies business ethics—and perhaps remind others of how a toilet operates.

What is all this AI doing? Promoters will point to many innovations—faster screening for cancer cells, closer connection to far-flung relatives, precision application of fertilizers and pesticides, elimination of drudgery in the workplace through automation. A bright future indeed—or perhaps not?

The real issue is whether or not AI data centers are economically viable, socially appropriate, environmentally sustainable, and actually serve the public interest.

In August 2025, ProPublica reported that the Food and Drug Administration (FDA) had lost 20% of its staff devoted to food safety thanks to DOGE cuts. Inspection of food import facilities is now at a historic low even as our dependence on the rest of the world to feed us grows. But not to worry, the FDA announced in May that AI was coming to the rescue thanks to a large language model (LLM)—dubbed Elsa—that would be deployed alongside what’s left of its human staff to expedite their oversight work. Hopefully, Elsa knows melamine when it sees it. AI chatbots are also growing in popularity and available 24-7 to “talk or advise” people on all sorts of pressing issues—how to win more friends, how to cheat on this exam, how to make up fake legal opinions, even encouraging a teenager to commit suicide and suggesting to someone else that they murder their own parents.

But there is an even dirtier AI underbelly. Some have dubbed these AI slop, AI smut, and AI stazi—three 21st-century horsemen of the digital apocalypse. What is this all about? Well, a lot of these accelerating AI algorithms are actually devoted to selling “products” that many people do not want and would find objectionable, as well as providing “services” that undermine our basic freedoms. Slop (Merriam Webster’s word of 2025) is used to describe when AI generates internet content that is only meant to make money through advertising. Right now there are thousands of wannabe internet “creatives” all over the globe, watching “how-to videos” to manufacture AI social media to grab the eyeballs of US consumers. That cute puppy video you see on Instagram or that shocking “news” story you read on Facebook is not by accident—the goal is to monetize clicks per thousand (cost per mille, or CPM) where advertisers pay for how much their ad is viewed online. This is also why online content is often overly long (where is the actual recipe in this cooking blog?), since that increases ad scrolling. The average US consumer is now subject to between 6,000 and 10,000 ads per day—70% of which are online. For more on AI slop, visit: https://www.visibrain.com/blog/ai-slop-social-media.

An even worse virtual commodity is AI smut—literally algorithms creating pornography. This perverted version of AI scraps the internet for images (high school yearbooks, red carpet fashion shows, popular music concerts, street cam footage, etc.) and then uses “face swap” programs to create personalized hardcore rubbish. There is little if any accountability for this theft of public images and violation of personal privacy—at best those involved are “shamed” into taking down their AI sites after being exposed due to fears of liability and prosecution for child abuse. But that has hardly stopped this seedy AI subsector. Can you imagine your face or image being put into such a lucrative sexploitative scenario without your permission? At this point, there are hardly any internet police walking the beat in the virtual AI world. We don’t even have the right to be forgotten on the internet.

Which brings us to AI stazi—the updated version of the Cold War-era East German secret police. University of Wisconsin Madison just announced the creation of a College of Computing and Artificial Intelligence (CAI), in part thanks to a $140 million donation from Cisco. Few Bucky Badger fans know that 30 years ago they were used as guinea pigs while cheering at Camp Randall Stadium to help create facial recognition technology through a UW-Madison grant from the Department of Defense Applied Research Agency (DARPA). Visitors to the UW campus today will no doubt “enjoy” the automated license plate readers (ALRPs) owned by Flock Safety. According to an August 2025 Wisconsin Examiner expose, there are hundreds of Flock cameras across the state in use by law enforcement agencies, including Wisconsin county sheriff departments with active 287(g) cooperation agreements with Immigration and Customs Enforcement. No warrant is needed for law enforcement agencies to browse the national Flock database. In fact, agents have used Flock to track peaceful protesters, spy on spouses, or just stalk people they don’t like. To see where Flock cameras are near you, visit: www.deflock.me. Of course, Flock Security has outsourced its AI programming to cheaper (and more secure?) Filipino contractors. Similar AI spying networks such as Pegasus have been widely exposed and have become “bread and butter” for authoritarian regimes from Israel to Saudi Arabia. China and Russia have their own versions (Skynet, SORM, etc.). Thanks to the cozy relationship between Trump and Peter Thiel, the US-based AI mercenary outfit, Palantir, is now being redeployed for domestic surveillance—first revealed by Edward Snowden back in 2017.

The latest executive bluster from Trump is that states’ rights are out the window when it comes to regulating AI data centers—such federal preemption of local democratic control is part of the larger neoliberal “race to the bottom” forced-trade agenda. But the cat is already out of the bag as dozens of communities have successfully blocked AI data center projects and others are poised to do the same based upon their winning strategies. Better yet, this is a bipartisan grassroots organizing issue!

What is the best way to keep out an AI factory farm? No non-disclosure agreements (NDAs)! These are massive development schemes that could not exist without the approval and support of elected officials, so any agreement should not be secret. They can hardly claim to be providing a public good if they are not subject to transparency and oversight. No sweetheart deals! Big Data is among the wealthiest sectors of our current economy and does not need or deserve subsidies, discounted electric rates, tax increment financing, property tax holidays, or other incentives. It is a classic move of crony capitalism to privatize the benefits and socialize the costs. No regulatory loopholes! Given their huge demands for land, water, and energy, Big Data should not be allowed to cut legal corners and needs to follow all the rules of any other normal enterprise—full liability coverage, no special economic zones, consideration of cumulative impacts, protections for ratepayers, no unregulated toxic pollution or illegal water transfer in violation of the Clean Water Act or the Great Lakes Compact, etc. How much water your data center demands is hardly a “trade secret.”

And most important, don’t let Big Data boosters belittle your legitimate concerns as “neo-Luddite!” Everyone uses technology—even the Amish. The real issue is whether or not AI data centers are economically viable, socially appropriate, environmentally sustainable, and actually serve the public interest. People have good reasons to be wary and oppose them on all those fronts.

For more info, checkout: Big Tech Unchecked: A Toolkit for Community Action

As well as the North Star Data Center Policy Toolkit


Our work is licensed under Creative Commons (CC BY-NC-ND 3.0). Feel free to republish and share widely.


John Peck
John E. Peck is the executive director of Family Farm Defenders.

 

Microbes in breast milk help populate infant gut microbiomes

University of Chicago Medical Center

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

 

image: 

Illustration by Pamela Ferretti

view more 

Credit: Illustration by Pamela Ferretti

Most conversations about breast milk tend to focus on topics like nutrients, antibodies and bonding time rather than bacteria. But it turns out that human milk carries its own tiny community of microbes, and those passengers may help shape a baby’s developing gut microbiome — which in turn can impact nutrient absorption, metabolic regulation, immune system development, and more.

A new study published in Nature Communications provides one of the most detailed portraits yet of how different combinations of bacteria in human milk contribute to the assembly of infants’ gut microbiomes.

Mapping the milk microbiome

The breast milk microbiome is notoriously difficult to analyze because the milk’s high fat content and relatively low bacterial load complicate the process of extracting genomic material.

“Breast milk is the recommended sole source of nutrition for an infant’s first months of life, but important questions about the milk microbiome remained unanswered because the analytical challenges are intimidating,” said first author Pamela Ferretti, PhD, a postdoctoral researcher in the Blekhman Lab at the University of Chicago. “We decided to tackle this endeavor because our collaboration presented a unique opportunity to combine key resources.”

Those resources included hundreds of milk samples collected as part of the Mothers and Infants LinKed for Healthy Growth (MILk) study, led by Ellen Demerath, PhD, at the University of Minnesota and by David Fields, PhD, at the Oklahoma University Health Sciences Center. On the UChicago side, Ferretti and her colleagues had access to metagenomic tools and deep experience with microbiome data, including Ferretti’s highly specific expertise in infant microbiomes and transmission analysis. In her previous research, she studied how different maternal body sites — such as mouth, skin, and vaginal cavity — contributed to infant microbiomes.

Analyzing 507 breast milk and infant stool samples from 195 mother–infant pairs, the team found that breast milk contained a distinct mix of bacterial species dominated by the genus bifidobacteria, including Bifidobacterium longum, B. breve, and B. bifidum. More than half of the milk samples carried B. longum, a species abundant in over 98% of the infants’ gut microbiomes.

“Even though B. longum is well-documented as being highly prevalent in the infant gut, it was surprising to find such a strong signature of that species in the breast milk samples because previous milk studies mostly reported other bacterial taxa like Staphylococcus and Streptococcus,” Ferretti said. “We think these results will prompt some reevaluation in the field.”

Tracing microbes from milk to the infant gut

Most prior studies analyzing bacterial DNA in breast milk used a relatively inexpensive, fast technique called amplicon sequencing, which targets a limited number of predetermined genomic regions for each experiment. This method is good for efficiently identifying species within a mixed sample, but it leaves most of the bacterial genome unexamined.

“Metagenomic analysis is trickier and more complicated, but it really paid off because it allowed us to obtain information at the level of different bacterial strains — which is key, because that’s the only level where we could actually claim to know about transmission,” Ferretti said.

The paper reported 12 instances in which the same exact strain was found in a mother’s breast milk and in the gut of her infant, which is a very strong indication that the transmission happens vertically via breastfeeding.

Some of these shared strains were beneficial commensal species such as B. longum and B. bifidum, which help digest human milk sugars and support healthy gut development. Others, however, were pathobionts — microbes like E. coli and Klebsiella pneumoniae that can live harmlessly in healthy individuals but have the potential to cause infection under certain conditions. The authors note that all mothers and infants in the study were healthy, indicating that these species’ presence in milk does not inherently signal disease but rather reflects the microbial diversity that can be transferred during breastfeeding.

Interestingly, the team also saw specific strains of bacteria usually associated with the mouth — such as Streptococcus salivarius and Veillonella species — in milk samples. They realized this as potential evidence of “retrograde flow” during breastfeeding: as the baby feeds, tiny amounts of oral bacteria may travel back into the nipple and ducts and become part of the milk microbiome.

Expanding human milk research

Ferretti noted that the study not only sheds light on microbial transmission but also fills a major gap in available data for scientists studying early-life health.

“This study nearly doubled the number of metagenomic breast milk samples that are publicly available, and pairs them with extensive information on mothers’ health and lifestyle,” Ferretti said. “We’re hopeful that our findings and future analyses that use this dataset will really push the field forward.”

In subsequent studies, the researchers hope to take their analysis to the next level with a multi-omic approach, including analyzing metabolites like human milk oligosaccharides (HMOs) and examining the “exposome” of environmental factors like PFAS and antimicrobial resistance that can be passed along through milk.

“Ultimately, we’re interested in looking at longer health trajectories to see if factors in breast milk and early life are predictive of health outcomes later in life,” Ferretti said.

 

“Assembly of the infant gut microbiome and resistome are linked to bacterial strains in mother’s milk” was published early access in November 2025 in Nature Communications. Co-authors include Pamela Ferretti, Mattea Allert, Kelsey E. Johnson, Marco Rossi, Timothy Heisel, Sara Gonia, Dan Knights, David A. Fields, Frank W. Albert, Ellen W. Demerath, Cheryl A. Gale and Ran Blekhman.

---

Journal

Nature Communications

DOI

10.1038/s41467-025-66497-y 

Method of Research

Observational study

Subject of Research

Human tissue samples

Article Title

Assembly of the infant gut microbiome and resistome are linked to bacterial strains in mother’s milk

 

Portable biosensor may enable on-site PFAS detection

La Trobe University

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

A portable biosensor developed at La Trobe University may allow rapid, on-site detection of toxic “forever chemicals” in water, removing the need for samples to be sent to specialist laboratories.

The device is designed to detect per- and polyfluoroalkyl substances (PFAS), a group of more than 15,000 synthetic chemicals used in products such as firefighting foams, food packaging and stain-resistant fabrics.

PFAS are highly persistent in the environment and have been linked to serious health risks, including cancer. Specifically, the sensing device detects PFOA (perfluorooctanoic acid), which is among the most regulated PFAS.

Led by PhD student Henry Bellette and Dr Saimon Moraes Silva, Director of La Trobe’s Biomedical and Environmental Sensor Technology (BEST) Research Centre, the research has been published in the journal ACS Sensors.

Bellette said current PFAS-testing methods limited how often and where water could be monitored.

“Most PFAS testing relies on expensive laboratory equipment and specialist analysis, which makes regular monitoring difficult,” he said. “This biosensor could be used on site and provides a simple yes or no result, allowing water to be screened quickly and easily.”

PFAS contamination has been identified across Australia, particularly at sites where firefighting foams were historically used, including airports, military bases and fire stations.

Dr Moraes Silva said portable screening tools would play an important role in managing long - term contamination.

“PFAS do not break down in the environment, so monitoring is an ongoing challenge,” he said. “A portable screening tool could allow more frequent testing, particularly in regional and remote areas, and help identify where more detailed laboratory analysis is needed.”

The researchers hope the technology can eventually be incorporated into a hand -held device for environmental monitoring and water screening.

DOI: https://doi.org/10.1021/acssensors.5c03678

---

Journal

ACS Sensors

DOI

10.1021/acssensors.5c03678 

Method of Research

Experimental study

Subject of Research

Not applicable

Article Title

Point-of-Need PFAS Detection: A Yes/No Biosensor SolutionArticle link copied!

‘Forever chemicals’ may increase liver disease risk in adolescents by as much as 3-fold

A research collaboration co-led by USC and the University of Hawai’i found that higher levels of two common types of PFAS in the blood were linked to an increased risk of early onset of MASLD, formerly known as fatty liver disease.

Keck School of Medicine of USC

FacebookXLinkedInWeChatBlueskyMessageWhatsAppEmail

A new study co-led by the Southern California Superfund Research and Training Program for PFAS Assessment, Remediation and Prevention (ShARP) Center and the University of Hawai‘i has linked certain common “forever chemicals” to a higher risk of liver disease in adolescents. These synthetic compounds, known as per- and polyfluoroalkyl substances (PFAS), may as much as triple the chances that adolescents develop a liver condition called metabolic dysfunction-associated steatotic liver disease (MASLD) — formerly known as fatty liver disease. 

The findings were published in the journal Environmental Research.

MASLD affects about 10% of children and up to 40% of children with obesity. It is a chronic condition that doesn’t always have telltale symptoms, although some patients experience fatigue, discomfort and abdominal pain. The disease increases long-term risk for type 2 diabetes, heart disease, advanced liver injury, cirrhosis and even liver cancer. 

“MASLD can progress silently for years before causing serious health problems,” said Lida Chatzi, MD, PhD, a professor of population and public health sciences and pediatrics and the director of the ShARP Center, a national center funded by the National Institute of Environmental Health Sciences to investigate PFAS health impacts, advance cleanup technologies and support affected communities. “When liver fat starts accumulating in adolescence, it may set the stage for a lifetime of metabolic and liver health challenges. If we reduce PFAS exposure early, we may help prevent liver disease later. That’s a powerful public-health opportunity.”

PFAS are manufactured chemicals used in nonstick cookware, stain- and water-repellent fabrics, food packaging and some cleaning products. They persist in the environment and accumulate in the body over time. More than 99% of people in the U.S. have measurable PFAS in their blood, and at least one PFAS is present in roughly half of U.S. drinking water supplies.

“Adolescents are particularly more vulnerable to the health effects of PFAS as it is a critical period of development and growth,” said the study’s first and corresponding author Shiwen “Sherlock” Li, PhD, an assistant professor of public health sciences at the University of Hawai‘i. “In addition to liver disease, PFAS exposure has been associated with a range of adverse health outcomes, including several types of cancer.”

Linking PFAS, genetics, and lifestyle

The research examined 284 Southern California adolescents and young adults from two USC longitudinal studies. The participants were already at higher metabolic risk because their parents had type 2 diabetes or were overweight. PFAS levels were measured through blood tests, and liver fat was assessed using MRI.

Higher blood levels of two common PFAS — perfluorooctanoic acid (PFOA) and perfluoroheptanoic acid (PFHpA) — were linked to a greater likelihood of MASLD. Adolescents with twice as much PFOA in their blood were nearly three times more likely to have MASLD. The risk was even higher for those with a genetic variant (PNPLA3 GG) known to influence liver fat. In young adults, smoking further amplified PFAS-related liver impacts.

“These findings suggest that PFAS exposures, genetics and lifestyle factors work together to influence who has greater risk of developing MASLD as a function of your life stage,” said Max Aung, PhD, MPH, assistant professor of population and public health sciences at the Keck School of Medicine. “Understanding gene and environment interactions can help advance precision environmental health for MASLD.”

Li noted that this study is the first to examine PFAS and MASLD in children using gold-standard diagnostic criteria, and the first to explore how genetic and lifestyle factors may interact with PFAS exposure. MASLD also became more common as adolescents grew older, adding to evidence that puberty and early adulthood may increase susceptibility to environmental exposures.

The study builds on recent USC research showing that, for adolescents undergoing bariatric surgery to manage obesity, a PFAS known as PFHpA is linked to more severe liver disease, including inflammation and scarring of connective tissue called fibrosis. 

“Taken together, the two studies show that PFAS exposures not only disrupt liver biology but also translate into real liver disease risk in youth,” Chatzi said. “Adolescence seems to be a critical window of susceptibility, suggesting PFAS exposure may matter most when the liver is still developing.”

About this study

Other co-authors are Jiawen Carmen Chen, Jesse Goodrich, Lily Dara, Lucy Golden-Mason, Ana Maretti-Mira, Zhanghua Chen, Frank Gilliland, Brittney Baumert, Sarah Rock, Sandrah Eckel, David Conti and Rob McConnell, all of USC; Elizabeth Costello, who is affiliated with USC and Brown University; Douglas Walker of Emory University; Scott Bartell and Veronica Vieira of UC Irvine; Tanya Alderete of Johns Hopkins University; Michael Goran of Children’s Hospital Los Angeles; and Alan Ducatman of West Virginia University.

The study was funded by the National Institutes of Health [P42ES036506, R01DK59211, 5P01ES022845-03, 5P30ES007048, 5P01ES011627, R01/ES029944, R01ES030691, R01ES030364, R01ES033688, U01HG013288, R01ES035035, R01ES035056, P50MD17344, T32-ES013678, ES035035], the U.S. Environmental Protection Agency [RD83544101] and the Hastings Foundation. 

---

Journal

Environmental Research

DOI

10.1016/j.envres.2025.123320 

Method of Research

Data/statistical analysis

Subject of Research

People

Article Title

Associations between per- and polyfluoroalkyl substances and metabolic dysfunction-associated steatotic liver disease in adolescents and young adults: modifying roles of age, lifestyle factors, and PNPLA3 genotype

Article Publication Date

12-Nov-2025

COI Statement

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Dr Chatzi and Dr Ducatman have served as an expert consultants for plaintiffs in litigation related to PFAS-contaminated drinking water.