PFOS: The Forever Chemical That Outlasted Its Own Phase-Out

3M stopped making PFOS in 2002. The chemical is still in nearly everyone tested. The carbon-fluorine bond that made it effective at repelling stains is the same bond that makes it permanent in the body — a half-life of nearly five years. This is the PFASPer- and polyfluoroalkyl substances — a family of over 14,000 synthetic chemicals built around the carbon-fluorine bond. that our endocrine disruptors guide keeps coming back to, and the one 3M spent decades measuring in its own workers before the rest of us found out.

PFOS — perfluorooctanesulfonic acid — is the eight-carbon PFAS that 3M industrialised in 1949 via electrochemical fluorination of POSF (perfluorooctanesulfonyl fluoride) and went on manufacturing until 2002. The most rigorous global production inventory puts 96,000 tonnes of POSF — PFOS's precursor — produced globally between 1970 and 2002, of which 3M was the dominant manufacturer (Paul, Jones & Sweetman 2009) of POSF made worldwide between 1970 and 2002, of which 3M was the dominant single manufacturer Paul et al. 2009. The chemical was the active ingredient in the original Scotchgard formula and in AFFFAqueous film-forming foam — firefighting foam historically formulated with PFOS-based surfactants. Used by military, airports, and petrochemical facilities since the 1970s. military firefighting foam. What made PFOS work — a fully fluorinated eight-carbon chain with a sulfonate head group — is what makes it almost impossible for your body to clear.

The sulfonate matters. Where its cousin PFOAPerfluorooctanoic acid — a carboxylate PFAS, IARC Group 1 carcinogen since 2023. The chemical behind the DuPont/Parkersburg contamination. carries a carboxylate group, PFOS carries a sulfonate (-SO₃⁻) that binds far more tightly to serum albumin. The result is a longer serum elimination half-life: 4.8 years geometric mean for PFOS — vs 3.5 years for PFOA — measured in retired 3M fluorochemical workers versus 3.5 for PFOA Olsen et al. 2007. Half the PFOS in your blood right now will still be there in 2031. The chemical designed not to wash off a sofa doesn't wash out of you either.

How far PFOS had spread shows up in the published record from 2001, when 3M chemist Kris Hansen and her colleagues Clemen, Ellefson and Johnson laid out a method for characterising fluorochemicals in biological matrices and applied it to 65 human sera samples obtained from commercial biological-supply companies. The donors were nonindustrially exposed. The samples were not supposed to contain the company's chemicals. They all did. The summed-fluorochemical totals matched the unidentified organic fluorine that had already been reported in human sera as far back as the 1970s — a signal nobody could name at the time Hansen et al. 2001.

Hansen's work didn't come out of nowhere. 3M's own peer-reviewed plant-worker biomonitoring at its Decatur, Alabama and Cottage Grove, Minnesota fluorochemical plants had been on the public record since 1980, in a paper by Ubel, Sorenson and Roach in the American Industrial Hygiene Association Journal Ubel et al. 1980. By the time the company announced its voluntary phase-out on 16 May 2000 — under the threat of EPA action that hadn't yet materialised — PFOS had already been documented in commercial-supply human sera, in occupational cohorts at 3M's own plants, and in pooled wildlife samples from the Great Lakes to the Arctic. The data weren't new in 2000. The decision was.

PFOS is permanent because of one bond. The carbon-fluorine bond running along its perfluorinated tail has a dissociation energy in the region of 485 kJ/mol carbon-fluorine bond dissociation energy — among the strongest single bonds in organic chemistry, around 40% stronger than the carbon-carbon bonds it sits between — among the strongest single bonds in organic chemistry, roughly 40% stronger than the carbon-carbon bonds holding the rest of the molecule together O'Hagan 2008. That strength is what makes PFOS useful for industrial coatings and what makes it impossible for the human body to dismantle. Bacteria don't break it. UV doesn't break it. Liver enzymes don't break it. The molecule that's in your serum today is the same molecule that left a 3M production line in 1985.

At the receptor level, PFOS does several things at once. It activates PPARαPeroxisome proliferator-activated receptor alpha — a nuclear receptor that regulates fatty-acid metabolism in the liver. Strongly activated in rodent liver by perfluoroalkyl acids; activated less potently in humans., the nuclear receptor that controls fatty-acid handling in the liver — but the rodent version of PPARα is activated far more potently than the human version, which is one reason rodent hepatotoxicity studies don't translate cleanly to human cancer risk Wolf et al. 2008 Lau et al. 2007. It competes with thyroxine for binding sites on transthyretinA transport protein that carries thyroid hormone and retinol in blood. PFOS competes with T4 for binding, with weaker affinity than T4 itself., the carrier protein that ferries thyroid hormone in the bloodstream, with weaker affinity than T4 — but with a high enough body burden to displace some bound hormone Weiss et al. 2009. And it suppresses the immune response in ways that are broadly independent of PPARα — which is why the immunotoxicity findings in humans replicate even when the rodent-to-human extrapolation breaks down for liver effects DeWitt et al. 2012.

This split matters for how the regulatory case got built. The mechanistic story for PFOS-and-cancer rests largely on rodent liver tumours driven through PPARα — a pathway humans share but engage less strongly. The mechanistic story for PFOS-and-immunity doesn't depend on PPARα at all, which is why the EFSA panel landed on it as the most sensitive endpoint when it set its 2020 tolerable weekly intake — and why one regulator can call PFOS a possible carcinogen on weak evidence while another sets the tightest dietary limit in the world based on a vaccine-antibody study in 101 babies.

PFOS production stopped at 3M in 2002 and was restricted internationally under the Stockholm ConventionAn international environmental treaty regulating persistent organic pollutants — chemicals that bioaccumulate, don't break down, and are transported across international boundaries. in 2009. The chemical is still detectable in more than 96% of people tested, because the carbon-fluorine bond that made it useful is the same bond that makes it permanent in groundwater, soil, and the food chain.

The dominant continuing source for most people is water. AFFF was sprayed at fire training grounds and crash sites for decades, and the PFOS went straight into the soil and from there into the aquifer. By the most recent GAO reporting, the US Department of Defense has identified 718 military installations where PFAS was used or may have been released — GAO-25-107401, June 2024 data where PFAS was used or may have been released, with preliminary assessment or site inspection completed at 712 of them. Freshwater fish are the other under-recognised route — PFOS biomagnifies up the aquatic food chain, and locally caught fish can carry concentrations far above what's in the water they swim in. The foam put out the fire. The PFOS went into the aquifer.

PFOS turned up in fish, birds, and marine mammals on every continent when researchers surveyed global wildlife in 2001. The team collected tissue samples from the Great Lakes to remote Arctic marine habitats — industrialised coastlines and pristine environments alike. The chemical was in virtually every sample, at concentrations that climbed at each step up the food chain Giesy and Kannan 2001. Animals that had never been near a factory or a fire training ground carried measurable PFOS in their tissue. Subsequent polar bear sampling across seven circumpolar populations found concentrations in some subpopulations exceeding immunotoxic effect thresholds Smithwick et al. 2005.

In humans, the trend is better. NHANES data shows that average serum PFOS in Americans has dropped 87% geometric mean dropped from 30.3 μg/L in 1999-2000 to 3.93 μg/L by 2017-March 2020 — NHANES since the phase-out — from a geometric mean of 30.3 μg/L in 1999-2000 to 3.93 μg/L by 2017-March 2020 Calafat et al. 2007 Kato et al. 2011. Detection has nonetheless run between 99% and 100% of the population aged twelve and over across every NHANES cycle since 1999, and the rate of decline has slowed in the most recent rounds — suggesting ongoing exposure from legacy sources.

Prenatal PFOS exposure cut children's diphtheria antibody response by 39% at age 5 — Faroe Islands birth cohort, n=587 — the strongest evidence that PFOS suppresses the immune system during development. The study that produced that number recruited 587 pregnant women in the Faroe Islands, a North Atlantic archipelago where pilot whale meat is traditional and a concentrated source of PFOS. Researchers measured PFOS in the mothers' blood during pregnancy, then drew blood from the children at ages five and seven to check how well their immune systems had responded to routine tetanus and diphtheria vaccinations. The children with higher prenatal PFOS made fewer antibodies at every measurement. By age seven, the children with higher PFOS at age five had 2.38 times higher odds of falling below clinically protective tetanus/diphtheria antibody thresholds — Grandjean 2012, PFOS-specific OR (95% CI 0.89-6.35) the odds of falling below clinically protective antibody levels; the equivalent odds ratio measured for PFOA in the same children reached 4.20 (95% CI 1.54-11.44) Grandjean et al. 2012.

The finding has been replicated. A follow-up study showed the most vulnerable window was the first six months after birth, with 19-29% decreases in tetanus antibody concentrations per doubling of PFAS exposure in early infancy decreases in tetanus antibodies per doubling of PFAS exposure in early infancy Grandjean et al. 2017. A Norwegian birth cohort of 99 mother-child pairs found inverse associations between prenatal PFOS and anti-rubella antibodies at age three Granum et al. 2013. Abraham and colleagues confirmed the pattern in 101 German one-year-olds, though in that population the strongest signal was for PFOA rather than PFOS Abraham et al. 2020. EFSAEuropean Food Safety Authority — the EU agency responsible for food safety risk assessment. used these vaccine antibody studies as the basis for its 2020 PFAS tolerable weekly intake. The strongest evidence for PFOS harm isn't cancer. It's that children exposed in the womb responded less to childhood vaccines.

Beyond immunity, PFOS tracks with raised cholesterol. NHANES 2003-2004 data from 860 adults showed individuals in the highest PFOS quartile had total cholesterol 13.4 mg/dL higher than the lowest PFOS quartile — Nelson et al. 2010, NHANES, n=860 higher than those in the lowest Nelson et al. 2010. Thyroid associations exist but are weaker and less consistent than for PFOA — a meta-analysis of 12 studies found small but statistically significant correlations between PFOS and thyroid hormone levels, with the strongest effects at intermediate concentrations suggesting a non-linear response Kim et al. 2018.

Reproductive endpoints add a third strand. Young Danish men in the highest combined PFOS+PFOA exposure group had 6.2 million morphologically normal sperm in highest combined PFOS+PFOA group vs 15.5 million in the lowest — Joensen 2009, n=105 young Danish men morphologically normal sperm against 15.5 million in the lowest group — less than half the count, with the difference holding at p=0.030, in 105 military draft recruits split by serum tertile Joensen et al. 2009. A follow-up study of 169 adult sons whose mothers' serum had been profiled during pregnancy in the Aarhus Birth Cohort found prenatal PFOA more strongly associated with reduced adult sperm concentration than PFOS specifically, but the in-utero pattern of perfluoroalkyl exposure tracking with later reproductive markers held across both chemicals Vested et al. 2013.

For birth weight the picture is genuinely contested. Pooled meta-analyses do find prenatal PFOS associated with modestly lower birth weights Negri et al. 2017. But a physiologically-based pharmacokinetic model — built to test confounding by maternal GFRGlomerular filtration rate — a measure of how quickly the kidneys clear substances from blood. Mothers with higher GFR clear PFOS faster AND tend to deliver larger babies, creating a confounded association. — showed that roughly half of the observed PFOS-birthweight association may reflect reverse causality: mothers with higher kidney clearance lose PFOS faster from their blood and also tend to deliver bigger babies Verner et al. 2015. The effect is real. The causal interpretation is harder than the headline.

On 30 November 2023, IARCInternational Agency for Research on Cancer — the WHO body that classifies substances by their cancer-causing potential. Group 1 = carcinogenic, 2A = probably, 2B = possibly. classified PFOS as Group 2B — 'possibly carcinogenic to humans' — based on strong mechanistic evidence but inadequate human cancer data IARC Monograph 135. PFOA, by contrast, got Group 1 — 'carcinogenic to humans' — on the basis of sufficient evidence in experimental animals plus strong mechanistic evidence in exposed humans. The human cancer evidence itself was rated only 'limited' for kidney and testicular cancer; it was the animal data and the mechanistic case in humans that pushed the classification up. PFOS's mechanistic and human cancer evidence didn't reach the same threshold — partly because no large concentrated community was ever exposed to PFOS in isolation the way the C8 cohort was exposed to PFOA.

The reason PFOS sits at IARC Group 2B while PFOA sits at Group 1 isn't biology. It's geography. PFOA had Parkersburg — a single DuPont plant whose discharge contaminated drinking water for some 69,000 people, generating a uniquely large, well-exposed cohort that the C8 Science Panel followed for years. PFOS contamination, by contrast, was diffuse: a Scotchgard factory here, an air-force base there, AFFF training grounds spread across hundreds of installations. No single concentrated community served as the natural experiment that PFOA's case rests on.

The closest equivalent is Ronneby, in southern Sweden, where firefighting-foam runoff from the Kallinge F17 air force base contaminated the municipal water supply from at least the mid-1980s until the contamination was identified in 2013. After the source was switched off, 106 residents aged four to 83 were followed and their PFOS, PFHxS and PFOA levels tracked as they declined. PFOS came down with a mean half-life of 3.4 years in the Ronneby cohort, 95% CI 3.1-3.7 — Li 2018, n=106 (95% CI 3.1-3.7), notably shorter than the 4.8-year geometric mean Olsen had reported in retired 3M workers Li et al. 2018. Half-life depends on more than the chemical itself — exposure source, matrix and population physiology all shift it.

Health-outcome studies in the Ronneby cohort have produced a mixed picture. Excess thyroid disease was the obvious thing to look for, and what turned up was a modest, non-monotonic increase in hypothyroidism medication prescribing among women in the mid-period of high exposure — equivocal enough that the authors themselves treated it as compatible with chance Xu et al. 2019. No headline. No PFOS Parkersburg. The cleanest natural experiment we have produces an answer that doesn't quite resolve.

PFOS has the longest regulatory trail of any PFAS — listed by the Stockholm Convention in 2009, restricted across the EU since 2006, and subject to a 4-parts-per-trillion PFOS maximum contaminant level in US drinking water — roughly four drops in an Olympic swimming pool US drinking water Maximum Contaminant Level finalised in April 2024. 3M announced the voluntary phase-out on 16 May 2000 under the threat of EPAUnited States Environmental Protection Agency action, after the company's own monitoring showed PFOS was bioaccumulating in workers and wildlife. Global production at 3M ended by 2002.

The EPA's 2024 final rule set MCLs of 4 ppt for both PFOS and PFOA, plus 10 ppt for PFHxS, PFNA and HFPO-DA (GenX) and a Hazard Index limit for the four-way mixture. Original compliance was required by 2029. In May 2025, EPA proposed to keep the PFOS and PFOA MCLs at 4 ppt while rescinding the MCLs for the other three PFAS and the Hazard Index, and to extend compliance to 2031. The proposed rule is expected to be finalised in spring 2026 EPA 2024 PFAS NPDWR.

The EU was first to legislate: Directive 2006/122/EC (the 30th amendment to 76/769/EEC) restricted PFOS marketing and use across the bloc, and the substance is now regulated as a persistent organic pollutantA chemical that is toxic, persists in the environment, bioaccumulates in living organisms, and is transported across international boundaries — the criteria for listing under the Stockholm Convention. under EU Regulation 2019/1021. The UK retained PFOS restrictions post-Brexit under the UK POPs Regulation. The Stockholm Convention listed PFOS in Annex B (restriction) at COP-4 in 2009, with broad exemptions for AFFF, semiconductor manufacturing, metal plating, aviation hydraulic fluids and a handful of specialty applications. The COP-9 amendment in 2019 (Decision SC-9/4) tightened that list — banning PFOS use in firefighting-foam training exercises, removing the aviation-hydraulic and several other exemptions, and retaining only Class B installed firefighting systems and Latin American leaf-cutting-ant baits as acceptable purposes. The amendment entered force in December 2020.

In Europe, the dominant consumer-facing limit isn't a drinking-water number — it's a tolerable weekly intake. EFSA's CONTAM panel set a TWI of 4.4 ng/kg body weight per week — the EFSA 2020 limit for the SUM of four PFAS: PFOA, PFNA, PFHxS and PFOS bodyweight per week for the SUM of four PFAS combined — PFOA, PFNA, PFHxS and PFOS — derived from the vaccine-antibody work in German one-year-olds EFSA CONTAM Panel 2020. EFSA's own dietary exposure assessment found that a substantial fraction of the European population already exceeds it: teenager exceedance averaged 14% exceedance across studied countries across studied countries, with the figure reaching 23.8% in France and remaining as low as 1.3% in Spain. Mean lower-bound adult exposure exceeds the TWI by a factor of five; P95 by a factor of sixteen. The number is published. The food supply is not yet adjusted to it.

The unfinished business outside the food chain is military AFFF. Per NDAA FY2020 implementation, the US Department of Defense was required to discontinue AFFF use at installations from October 2024, with installation-specific waivers available until October 2026 and a continuing exemption for shipboard fire suppression. Training-use was halted earlier; fluorine-free alternatives are now specified under Navy MIL-PRF-32725. The shipboard exemption persists beyond 2026.

The wider European response is moving slowly. In January 2023, five member states — Germany, the Netherlands, Denmark, Sweden and Norway — filed a joint REACH Annex XV proposal to restrict the entire PFAS class. After 5,600 public comments and an updated proposal in August 2025 that added eight industry sectors and 74 derogated uses with phase-outs of up to 13.5 years, the file sits with ECHA's RAC and SEAC committees. SEAC's draft opinion was provisionally due in early 2026, with a final committee opinion expected by year-end and a final restriction regulation in 2027 at the earliest. The class restriction will cover thousands of PFAS compounds at once — but in the meantime, PFOS continues to be regulated under the existing POPs framework rather than as part of the broader class.

With a half-life of 4.8 years, PFOS doesn't respond to quick dietary changes the way BPA does — reducing exposure is a long game of preventing new intake rather than flushing out what's already there. The practical levers are water and food, and they vary by where you live — our PFAS guide covers cookware and food packaging, while the tap water guide compares filter options.

PFOS is what happens when a chemical is too useful to question and too persistent to take back. 3M's own plant-worker biomonitoring went into the public scientific record in 1980; commercial-supply human sera came back universally positive in the work that became Hansen et al. 2001; the company announced its phase-out in 2000. The production stopped. The molecule didn't.

Serum levels in the US population have dropped roughly 87% since the phase-out — a genuine public health success. But detection still runs at near-100% across the population aged twelve and over, the decline has slowed, and the legacy contamination at 718 DoD installations will take decades to remediate. The regulatory system eventually caught up: Stockholm Convention listing in 2009, COP-9 amendments in 2019, EFSA's tolerable weekly intake in 2020, EPA's drinking water MCL in 2024, and a broader REACH PFAS class restriction now grinding through ECHA committee review. The half-life of PFOS is 4.8 years. The regulatory response time was longer.