Seed Oils: A Novel Industrial Food Worth Minimising (Not the Proven Poison, Not the Proven Health Food)

Summary

Industrial seed oils are a genuinely novel food eaten at unprecedented volume (linoleic acid went from ~2% to ~7%+ of calories in a century, soybean oil up over a thousandfold), promoted by an establishment with a documented conflict of interest, to replace traditional fats that were demonised on a saturated-fat premise that has since weakened — so the honest, recovery-posture verdict is precautionary: minimise refined/industrial seed oils, especially repeatedly-fried ones, and don't fear traditional whole-food fats — while being equally honest that the loud "seed oils are toxic and inflammato

Why Moderate

Moderate. The precautionary recommendation (minimise refined/fried seed oils, prefer whole-food/traditional fats) is a defensible Moderate posture supported by the consumption-novelty data, the oxidation chemistry, the weakened SFA-replacement premise, and the failed-benefit RCTs. The strong harm claim (seed oils are toxic/inflammatory) stays explicitly unsupported (Johnson & Fritsche RCT-null; Marklund lower-mortality). Established: the consumption magnitude, that SFA raises LDL, that LA doesn't acutely inflame in RCTs, that fried/oxidised oils generate harmful aldehydes, that biomarker LA tracks lower mortality. Contested: whether habitual refined-LA intake harms long-term outcomes, the o6:o3 clinical weight, and the causal reading of the temporal data.

Sub-claim tiers, which do not all sit at the entry's level:
• Refining contaminants (3-MCPD and glycidyl esters): Strong. A public regulator characterised the hazard, ran an exposure assessment, named an at-risk group, and the EU wrote dated numeric limits into binding law. That is the highest-graded plank in this entry — but it grades a refining argument, not a seed-oil argument, and it is capped, declining, and flagged for formula-fed infants rather than adults.
• Hexane residue: Emerging, low confidence, precautionary only. It is graded low because it is genuinely low: an unresolved regulatory file, occurrence data the regulator itself calls insufficient, no demonstrated harm at dietary exposure, and a costless alternative. It is the weakest plank here and is ranked accordingly. Grading it low is not a statement that hexane has been shown safe — the honest position is that it has not been assessed on current standards either way, and will not be until late 2027.
• Dietary linoleic acid and skin/UV: split, and the split matters. The mechanism half — diet remodels skin lipids, UV peroxidises membrane polyunsaturates producing the same aldehydes as the fryer, and LA is structurally required in the epidermal barrier — is Foundational to Strong and genuinely true. The human causal chain — that dietary LA makes people burn faster, age faster, or develop skin cancer — is Emerging at most and honestly NOT SUPPORTED; there is no positive human signal to grade, only a well-powered null and two direct dissociations running the wrong way. We publish this because our own research went against our directional prior, which is the only reason to trust the rest of the entry.

Practical takeaway

The recovery-posture default: minimise the industrial form, don't fear traditional fats, prioritise whole foods.
• Cut the highest-risk exposure first: repeatedly-fried and deep-fried oils (restaurant fryers, fried fast food) and the ultra-processed packaged foods built on refined seed oil. This is where the oxidation chemistry and the worst outcomes actually concentrate.
• For home cooking, prefer whole-food and oxidatively-stable fats: extra-virgin olive oil, avocado, butter/ghee/tallow in moderation, nuts and seeds — over large amounts of refined, high-linoleic-acid oil. You do not need to fear butter or tallow; the saturated-fat-causes-heart-disease premise that demonised them has weakened.
• Don't panic-audit every label or treat liquid seed oil in a whole-food meal as poison — that's the overcorrection the human RCT and biomarker data don't support. A drizzle of canola in a home stir-fry is not the issue; the deep-fried, repeatedly-heated, ultra-processed exposure is.
• Ensure omega-3 adequacy from oily fish (omega3_repletion) rather than chasing the omega-6:omega-3 ratio as a number.
• Recovery framing: when a recent industrial food has replaced foods we evolved eating, and its safety isn't established, the precautionary default is to lean toward the older, whole-food fats — without claiming the new one is proven toxic.

Evidence detail

Why This Entry Exists

This is the loudest food fight on the internet, and the two camps are each half-right in a way that makes the truth genuinely hard to find. The influencer camp says seed oils are toxic, inflammatory, and the cause of modern disease; the debunker camp says all of that is pseudoscience and the topic is settled-safe. Neither is honest about the other's strongest evidence.

This entry exists to hold the genuinely-uncertain middle and to take the skeptic case seriously, because the precautionary version of it is more defensible than the debunkers allow: industrial seed oils really are a ~100-year-old novel food, the body that championed them really did have a conflict of interest, the traditional fats they replaced really were demonised on a now-weakened premise, the oxidation chemistry really is valid, and the randomised trials of swapping saturated fat for high-dose linoleic acid really did fail to show a survival benefit (one showed harm). And — held in the same hand — the claim that seed oils are acutely toxic or inflammatory in humans genuinely fails the controlled trials, and higher tissue linoleic acid tracks lower mortality in biomarker cohorts. So the defensible stance is precautionary minimisation of the industrial/refined/fried form, not a toxicology verdict. It pairs with dietary_fat_cholesterol_and_testosterone (the omega-6 oxidative-stress/hormonal angle) so Realised speaks with one consistent, skeptic-respecting voice.

What bad advice this protects against, both ways:
• "Seed oils are toxic, they're poisoning everyone" → asserting a human toxicity/inflammation claim the RCTs don't support, and missing that the real distinction is refined/fried vs whole-food and traditional fat.
• "Seed oils are proven healthy, eat them freely, the science is settled" → ignoring the novelty, the documented industry COI, the weakened saturated-fat premise that justified them, the oxidation chemistry, and the recovered-data trials that failed to show benefit.
• "Swap your butter/tallow for vegetable oil, it's heart-healthy" → resting on a saturated-fat-causes-heart-disease premise that large analyses have substantially weakened.

Evidence

The honest skeptic case (the part the debunkers under-weight):

1. This is a genuinely novel food, eaten at unprecedented volume (Strong Evidence for the magnitude). Blasbalg et al. (AJCN 2011, government-funded, no conflict) documents that US linoleic acid availability rose from ~2.2–2.8% to 7.2% of energy across the 20th century, and soybean-oil consumption rose ~1,163-fold (0.009 → 11.64 kg/capita/yr, 1909→1999). US adipose-tissue linoleic acid roughly doubled (9.1% → 21.5%, 1959–2008; DiNicolantonio & O'Keefe 2018). Pre-industrial fat came mostly from animal fats and whole foods. The honest framing is a burden-of-proof argument: when a refined industrial food substitutes at this scale for foods humans evolved eating, the burden is on demonstrating safety. (Limit: novelty is not harm, and whole-food omega-6 from nuts/seeds is ancient — the argument is specifically about refined, high-volume, extracted, often-oxidised oils.)

2. The establishment that promoted them had a documented conflict of interest (Strong Evidence for the history, Emerging for the inference). The modern American Heart Association was launched by a 1948 fundraising windfall from the "Walking Man" radio contest sponsored by Procter & Gamble's Duz — P&G being the maker of Crisco (hydrogenated cottonseed oil). (Accurate wording: P&G sponsored the program that drove the donations; the AHA disputes a direct corporate cheque — don't overstate it as a clean bribe.) The AHA then codified the diet-heart, PUFA-over-saturated-fat position early and re-endorsed it in 2017 (Sacks et al. Presidential Advisory), which was criticised for leaning on a few old trials and downplaying the recovered-data nulls. And industry capture of 1960s nutrition science is proven: Kearns et al. (JAMA Intern Med 2016) showed the sugar industry paid Harvard researchers to steer blame toward fat. (Limit: a conflict of interest doesn't falsify the lipid biology — it justifies scrutiny, not dismissal.)

3. The traditional fats seed oils replaced were demonised on a weakened premise (Strong to Moderate). The "swap butter/tallow for vegetable oil" logic rests on saturated fat causing heart disease — which large analyses have substantially undercut: Siri-Tarino (AJCN 2010, n=347,747: SFA not significantly associated with CHD, RR 1.07), Chowdhury (Annals 2014), and de Souza (BMJ 2015: SFA not associated with mortality or CVD — while trans fats were). So traditional animal fats are not the proven hazard the mid-century model claimed. (Limit: SFA does raise LDL, and Hooper 2020 Cochrane (n=53,758) found reducing SFA cut cardiovascular events 17%, best when replaced by PUFA — but with no mortality reduction. "Eat unlimited butter" overshoots too.)

4. The oxidation chemistry is real and reaches human plaque (Moderate to Emerging). Linoleic acid's double bonds make it oxidation-prone; refining and especially repeated high-heat frying generate aldehydes (4-HNE) and oxidised LA metabolites (OXLAMs: 9-/13-HODE). OXLAMs are found in human atherosclerotic plaque (Brooks 1970; Belkner 1991), reportedly 20–100× higher in atherosclerosis patients than controls, and LA's ~2-year adipose half-life means intake durably remodels tissue lipids (the "oxidised linoleic acid hypothesis," DiNicolantonio & O'Keefe, Open Heart 2018). (Limit: the severe harm data are heated-oil/animal/mechanistic; habitual non-fried human intake is the uncertain part.)

5. The recovered-data RCTs failed to show benefit — one showed harm (Strong Evidence, and they outrank the cohorts on design): Sydney Diet Heart (Ramsden BMJ 2013): replacing SFA with safflower-oil LA raised mortality (HR 1.62). Minnesota (Ramsden BMJ 2016, n=9,423): corn-oil LA lowered cholesterol 13.8% but produced no mortality benefit — cholesterol-lowering tracked higher death in the over-65s. (Limit: both old, high doses, secondary-prevention/institutional, and the Sydney margarines carried trans fat; they refute "swapping in seed oil is proven to save lives," not "LA is toxic.")

6. The temporal correlation is suggestive but heavily confounded (Emerging to Experimental). Rising LA intake (~2%→7% of energy, per Blasbalg above) roughly paralleled the 20th-century rises in obesity and metabolic disease. But this cannot carry causal weight: sugar, ultra-processed food, total calories, smoking and sedentariness all changed together (and seed oils are components of UPFs), and CHD mortality has fallen since the 1960s–70s even as LA intake kept climbing. Hypothesis-generating only.

7. Refining creates contaminants that were not in the seed, and regulators have capped them in law (Strong Evidence — the best-documented processing-novelty plank in this entry). Industrial deodorisation runs oil at 220–260 °C under vacuum, and that step generates two contaminant families absent from the crude oil: 3-MCPD esters and glycidyl esters. EFSA's 2016 CONTAM opinion assessed 3-MCPD, 2-MCPD and glycidyl esters across 7,175 occurrence data points and derived a benchmark-dose lower confidence limit for 3-MCPD of 0.077 mg/kg bw/day for renal tubular hyperplasia in rats (a figure the 2018 re-evaluation replaced with 0.20 mg/kg bw/day; the 2016 derivation is retained here only to describe what the original opinion assessed, and 2-MCPD toxicity data were explicitly judged insufficient for a dose–response). European regulators class glycidol as genotoxic and carcinogenic (Regulation (EU) 2018/290, recital 5) and manage it by margin of exposure — on the pre-mitigation occurrence data available in 2016, T25 of 10.2 mg/kg bw/day gave margins of 11,300–102,000 at mean intake and 4,900–51,000 at the 95th percentile, against EFSA's low-concern threshold of 25,000. Those margins describe the market before FEDIOL's 2017 cap and the 2018 and 2021 Regulations; they are the reason the limits were written, not a description of oils on a European shelf today. The EU then wrote binding maxima. Regulation (EU) 2018/290 (in force 19 March 2018) caps glycidyl esters at 1,000 µg/kg in vegetable oils and fats sold to the final consumer or used as an ingredient, 500 µg/kg in oils for baby food, and 50 µg/kg in infant formula powder (6.0 µg/kg liquid) from 1 July 2019; its recital 4 records that these are the contaminants "found at highest levels in refined vegetable oils and fats." Regulation (EU) 2020/1322 (applying 1 January 2021) caps 3-MCPD and its esters at 1,250 µg/kg for coconut, maize, rapeseed, sunflower, soybean, palm-kernel and refined olive oils, 2,500 µg/kg for other vegetable oils including palm (and fish and marine oils), 750 µg/kg for baby-food oils, and 125 µg/kg in infant formula powder (15 µg/kg liquid). Formation is not uniform: glycidyl esters are strongly temperature-driven (sharply increased above 240 °C, only low sub-limit concentrations at 210 and 230 °C, largely avoidable below 230 °C), whereas 3-MCPD and 2-MCPD esters are barely temperature-dependent and are already present at the mildest condition tested (210 °C for 30 minutes, with potential formation above 180 °C).

Four honesty clauses travel with this plank and must not be dropped:
• The tolerable intake was loosened, not tightened. EFSA revised the 3-MCPD group TDI upward in 2018 to 2 µg/kg bw/day (BMDL10 0.20 mg/kg bw/day in male rats), and Regulation (EU) 2020/1322 recital 3 cites that figure. The older, stricter 2016 number is superseded and must never be quoted — quoting it would be exactly the selective citation this entry exists to refuse.
• The flagged group was infants, not adults. EFSA did not find general adult dietary exposure concerning. What it flagged was "slight exceedance of the TDI … in case of the high consumers of the younger age groups and in particular in case of infants receiving formula only" (2.4 µg/kg bw/day mean, 3.2 at the 95th percentile, against the 2 µg/kg TDI). A margin of exposure below EFSA's threshold is a risk-management trigger, not a demonstration of harm, and nothing here establishes harm to adults at dietary exposure.
• This is a refining argument, not a seed-oil argument. Palm — a fruit oil — is consistently the worst oil for both contaminants, attributed to its natural 4–12% diacylglycerol content (the glycidyl ester precursor); refined fruit oils such as palm and olive generate higher 3-MCPD esters than seed oils such as rapeseed or maize; and the law encodes the same ranking, putting seed oils in the stricter bucket, palm in the looser one, and exempting virgin olive oil because it is not refined. Chemically and legally, seed oils are the better-behaved category here, and a both-ways honesty doctrine has to say so out loud.
• Levels have fallen, and industry moved before the regulator did. FEDIOL, the EU vegetable-oil industry association, committed in mid-2015 to cap glycidyl esters in all food-use vegetable oils at 1 mg/kg by September 2017 — two and a half years before any law required it — and the 2018 Regulation then adopted that exact number. Independent confirmation of decline comes from Health Canada, which analysed infant formula on the Canadian market in 2015 (35 samples) and 2019 (33 samples): bound glycidol in reconstituted formula fell from a mean of 8.7 ng/g (max 40.3) to a mean of 6.7 ng/g (max 31.5) — a genuine but modest decline of roughly 23% on the mean. Effective mitigations include crude palm oil washing, non-hydrochloric-acid-activated bleaching earth, dual-temperature deodorisation and double refining. The correct tense is past-and-improving, not present-and-worsening.

And the popular version of this plank — "refined oil is already oxidised before you open the bottle" — is false and we do not print it. The stagewise refining measurements run the other way: rapeseed peroxide value falls from crude (3.57–5.40 mmol/kg) through bleaching (3.17–5.01) and falls further after deodorisation; p-anisidine value falls from 2.57–1.71 in crude to 1.78–0.99 in fully refined oil; freshly produced refined oils usually sit well below a peroxide value of 1.0 meq/kg; and cold-pressed rapeseed is reported to be more susceptible to autoxidation than refined, because unrefined oil carries more hydroperoxides and free fatty acids to begin with. Deodorisation is a stripping step — it removes volatile oxidation products. What survives as a real cost is different and worth stating precisely: refining removes roughly 50–60% of the oil's own tocopherols (deodorisation alone 32.8–42.2%), stripping its antioxidant reserve, and trans isomers absent at 190–210 °C increase significantly at 250–270 °C. The defensible sentence is antioxidant-depleted and chemically altered, not pre-oxidised. (Declared gap: every oxidation measurement located here was on freshly refined oil. Whether refined and cold-pressed oils diverge at retail, after months of shelf storage, is unverified in both directions — mechanistically plausible given the stripped tocopherols, but we assert nothing either way.)

8. Hexane extraction residue is an OPEN REGULATORY QUESTION — not a demonstrated harm, and the weakest plank in this entry (Emerging, low confidence, precautionary only). Most refined seed oil is extracted with hexane, a petroleum solvent, and residues are near-universal in finished oil. What follows is a precautionary argument about an unresolved safety file, and it must never be read as a harm claim.
• What is regulated. The EU sets a maximum residue of 1 mg/kg in the fat or oil under Directive 2009/32/EC, Annex I Part II. That Directive defines the substance not as pure n-hexane but as "a commercial product consisting essentially of acyclic saturated hydrocarbons containing six carbon atoms and distilling between 64 °C and 70 °C" — a technical mixture, which is precisely why impurity specification later became a data gap. The FDA sets no residue limit for hexane in vegetable oils and does not require it on the label, because it is regulated as a processing aid rather than an ingredient (incidental additives, 21 CFR 101.100(a)(3)); the only US numeric limits sit in narrow, unrelated categories — 25 ppm in spice oleoresins (21 CFR 173.270) and 5 ppm in fish protein isolate (21 CFR 172.340). Other jurisdictions are reported rather than verified here: Australia/New Zealand 0.3 mg/kg, Canada 10 ppm, and China 10 mg/kg — the last of these ten times looser than the EU, which cuts against any reading in which the EU limit is the permissive one.
• What actually gets measured. A market survey of 40 samples across 23 brands detected hexane in 36 of 40, from below the limit of detection to a maximum of 42.6 µg/kg — and its own conclusion is the decisive part: every sample was below the EU limit, the maximum being about 4% of it. A separate method paper (n=4 bottles) found one exceedance, a sunflower oil at 2.688 mg/kg. These two cannot be combined into a prevalence figure — one found zero exceedances in forty samples, the other one in four — and we do not combine them. The honest statement is: residue is routinely present and detectable, typically one to two orders of magnitude below the legal limit, and occasional exceedances do occur.
• What the hazard classification does and does not say. n-Hexane carries an EU harmonised classification as a suspected reproductive toxicant (Category 2) and for repeated-exposure organ toxicity. Neither is a statement about a trace residue in food. Hazard classification describes intrinsic properties of the substance in bulk; Category 2 is the suspected tier, not the known or presumed one; and the repeated-exposure classification carries a concentration limit reported as 5% hexane (unverified against Annex VI — see the verify note), which would put refined oil at the EU residue limit — 0.0001% hexane, roughly fifty thousand times below — far beneath the point at which the classification applies to a mixture at all. Note that this is a labelling cut-off, not a dose–response threshold. The well-documented n-hexane peripheral neuropathy is an occupational inhalation phenomenon mediated by the metabolite 2,5-hexanedione, seen in shoe, furniture and printing workers at airborne concentrations sustained over months. It is not a food finding and must never be presented as one.
• Why it is nonetheless open. The safety basis is old. The 1996 Scientific Committee on Food set an oral no-observed-effect level of 23 mg/kg bw/day against an estimated dietary exposure of 0.1 mg/kg bw/day, a margin of safety around 200. EFSA is now re-opening it at the Commission's request: its 2024 technical report concluded that "additional information on the identity, impurities and use of the solvent, on the actual occurrence in food and on toxicological data is needed," and its exposure assessment was built on the regulatory limits, not on measured residues, with a refined assessment possible only after occurrence data are collected. Under that maximum-limit ceiling scenario, exposure for infants below 16 weeks reached up to 7.8 mg/kg bw/day — against the no-effect level of 23, a margin of roughly 3. That figure is not comparable to the 1996 margin of around 200: the 1996 number was computed against estimated actual dietary exposure across the general population, whereas 7.8 mg/kg bw/day is a ceiling calculation assuming every food sits at its legal maximum, for the highest-exposure infant group. No margin has been revised and nothing has been re-measured — the point is only that the assessment has not been redone on current data. It must always travel with the other half: EFSA did not identify any immediate health concern linked to oral exposure to hexane. The call for data closed in December 2025 and the final risk assessment is due 22 November 2027.
• Why it still meets a precautionary trigger. The safety basis dates to 1996, the residue occurrence data the regulator itself calls insufficient, the file is formally re-opened until late 2027 — and no proven benefit is forgone by choosing cold-pressed or olive oil instead. (For butter or tallow the swap is not costless — the Cochrane analysis above found saturated-fat reduction cut cardiovascular events, best when replaced by PUFA, so that substitution trades an unresolved regulatory question for a demonstrated event-rate cost.) An unresolved question with a costless alternative is the textbook precautionary case. That is the whole of the argument. It is not a claim that hexane causes harm at dietary exposure, and this entry never makes one.
• The symmetry, stated plainly, because this plank's sourcing is contaminated on both sides. FEDIOL and EUFIC — the refining industry and its communications arm — defend hexane extraction, and their concession is narrow but real: they publicly acknowledge the re-assessment rather than denying it. Attacking it are cold-pressed and expeller-pressed producers, the ancestral-diet and tallow retail category that uses hexane as its central image, and — most importantly — manufacturers of competing solvents. Three of the source claims we tested on this plank traced back to firms selling the hexane replacement, and two of the three were wrong or untraceable: a US regulatory figure inflated tenfold in the direction that flattered the anti-hexane case, and a national limit that could not be found in any instrument of the country it was attributed to. Both errors leaned the same way. EFSA, by contrast, is the EU's public risk assessor, not an interested party — its re-assessment is stronger evidence for that reason, but it is not an "admission against interest" and we do not frame it as one. And Realised's own incentive belongs on this ledger: a platform positioned on honest evidence has a structural pull toward discovering that the mainstream is wrong, because an open question is more interesting than a closed one. That pull is exactly what produced the errors caught above. The rule it forces is the one we hold ourselves to here: when a figure flatters our argument, verify it against the primary text before printing it, and when the primary text says the opposite, print the opposite.

The hard guardrails (the part the skeptics under-weight — keep these honest):

9. Linoleic acid does NOT raise inflammatory markers in human trials (Strong Evidence). Johnson & Fritsche (2012), systematic review of RCTs in healthy adults: adding dietary LA produced no significant increase in CRP, IL-6, TNF-α, fibrinogen, or adhesion molecules. The "seed oils are acutely inflammatory in humans" claim is not supported and must not be asserted as fact.

**10. Higher blood/tissue linoleic acid tracks lower mortality (Strong Evidence, biomarker cohort).** Marklund (Circulation 2019, ~69,000 people, 30 cohorts): higher in-vivo LA was associated with lower CVD mortality (HR ~0.78) and stroke; arachidonic acid was not associated with higher risk. This is biomarker-based (not diet recall), and it cuts against "LA accumulates and harms" for habitual non-fried intake. (It is an association, not randomised proof of benefit.)

11. The "dietary linoleic acid makes you burn faster / causes skin cancer" claim does NOT hold — and we print that because our own research went against us (Emerging at most, and honestly presented as NOT SUPPORTED). This is the sub-claim where the evidence ran against this entry's own directional prior. It is included for exactly that reason.

The half that holds, and it is real: dietary fat composition genuinely does remodel skin lipids — one human study drove epidermal omega-3 from 1.8% to 24.2% of total fatty acids by diet alone. UV genuinely does peroxidise polyunsaturated membrane lipids, and the products include 4-HNE and malondialdehyde — the same aldehyde families as the fryer, which is a genuine mechanistic through-line to this entry's strongest plank. And linoleic acid is structurally required in the epidermal water-permeability barrier, as omega-O-acylceramide, LA esterified to ultra-long-chain ceramides; its absence produces the scaly, water-losing skin of essential fatty acid deficiency, described since the 1930s. That mechanistic half is Foundational to Strong, not weak evidence, and rating it low would hand an easy rebuttal to anyone who knows dermatology.

The half that fails, on four independent counts:
• The argument is not applied consistently, and condemns fish oil harder than seed oil. Peroxidisability scales with the number of bis-allylic methylenes: linoleic 1, alpha-linolenic 2, arachidonic 3, EPA 4, DHA 5. On the conventional peroxidisability index, DHA runs 5–8× linoleic acid. Anyone arguing "LA is dangerous because it peroxidises in skin" is committed to a much stronger warning about fish oil, which the same people recommend. (This is a consistency argument, not an outcome argument — it shows the popular reasoning is self-refuting; it does not show LA is safe. The obvious rebuttal, that typical LA intake is roughly two orders of magnitude above EPA+DHA intake so total substrate beats per-molecule rate, is closed off empirically by the next point.)
• Peroxidation and sunburn injury dissociate in the two studies that measured both — and in humans they run the wrong way. Takemura et al. (2002, hairless mice) fed linoleic-acid versus alpha-linolenic-acid diets: topically the two behaved alike, but on the dietary route **the ALA group had a significantly higher skin peroxidisability index and significantly lower UVB-induced prostaglandin E2, with no significant difference in actual lipid-damage markers** between groups. Rhodes et al. (1994, J Invest Dermatol) is the human version: 15 volunteers on 10 g/day fish oil raised their minimal erythema dose from 18.9 ± 13.9 to 41.1 ± 16.6 mJ/cm² at six months (p<0.01), falling back to 23.1 ± 4.9 ten weeks after stopping (p<0.05) — a dose-on/dose-off pattern that is unusually good internal evidence in a small study — while epidermal omega-3 rose 1.8% → 24.2% and skin peroxidation markers in irradiated skin rose from 6 ± 0.3 to 18.5 ± 2.6 (p<0.01). The authors' own reading is that omega-3 "may act as an oxidizable buffer, protecting more vital structures from free radical damage." In humans, roughly tripling skin lipid peroxidation more than doubled the sunburn threshold. That is the popular causal chain running backwards, in people, in the same tissue. (Limits: mice in the first; n=15, no placebo control, a crude non-specific peroxidation assay and an erythema rather than cancer endpoint in the second.)
• "LA in skin is bad" inverts an essential nutrient's role. LA is the primary epidermal omega-6 and is required for the barrier, as above. A 2024 cell study additionally found linoleic acid itself reduced UVB-induced reactive oxygen species in keratinocytes and restored UV-depleted glutathione — but that is an immortalised cell line with the fatty acid applied directly rather than incorporated by diet, published in a low-barrier venue, with a funding statement we could not retrieve. We flag it as mechanistically interesting and lean on it for nothing. This entry does not claim LA is protective.
• The human trials do not isolate linoleic acid and do not agree. Black et al. (NEJM 1994) is the one positive trial: 76 non-melanoma skin cancer patients, dietary fat cut from ~40% to 21% of energy, cumulative new actinic keratoses over months 4–24 of 10 ± 13 (control) vs 3 ± 7 (intervention), P = 0.001. The load-bearing objection is decisive — the intervention was total fat reduction, so nothing in it can be attributed to linoleic acid — and it is single-centre, participant-unblinded, on a premalignant surrogate endpoint, with standard deviations larger than the means. (A later report of reduced skin cancer occurrence is the same 76 patients extended, not independent replication.) Against it sits the Women's Health Initiative dietary-modification randomised trial: 48,835 postmenopausal women, 8.1 years, non-melanoma skin cancer HR 0.98 (95% CI 0.92–1.04), melanoma HR 1.04 (0.82–1.32) — publicly funded, no commercial sponsor, and a tightly-bounded null rather than an underpowered one, which is the strongest form a negative result can take. (Both-ways honesty requires the finding that embarrasses the low-fat side too: baseline fat intake interacted with assignment, p = 0.006, and among women with higher baseline fat intake, fat reduction was associated with increased melanoma risk, HR 1.48, 95% CI 1.06–2.07. That is a subgroup result inside a null trial and is hypothesis-generating only.) Observational epidemiology is predominantly null and inconsistent rather than cleanly negative — scattered subgroup associations exist.

The strongest single item is a concession by the hypothesis's own author. Black and Rhodes (2016) — the author of the 1994 trial and the animal photocarcinogenesis work that built the dietary-fat/skin-cancer hypothesis, writing with the author of the fish-oil study — concede that "for the most part, both case-control and prospective cohort studies have failed to find a relationship between skin cancer incidence with dietary fat intake," and concede the peroxidation paradox in their own words: the rise in minimal erythema dose on fish oil "was accompanied by an increase in epidermal omega-3 FA composition and increased susceptibility to lipid peroxidation." Two authors each conceding the limits of their own signature finding, in one paper, with charitable and public funding and no conflict declared. (Two precision points: they do not reference the Women's Health Initiative — do not say they conceded that null. And it is an admission against the fat-causes-skin-cancer hypothesis, not against seed oils specifically; the review's own thesis is still that omega-3 may benefit skin cancer.)

Named and quarantined: this claim propagates almost entirely through commercial ancestral-diet and tallow sellers, and its strongest-sounding datum — a higher polyunsaturated fat percentage in melanoma patients' tissue — is a cross-sectional tissue-composition comparison reachable only through blog citation, with reverse causation and confounding unhandled. Realised does not use it, in copy or in chat. Note the funding asymmetry cleanly and without gloating: every source that undermines this claim is academic, charitable or publicly funded, and the best-funded party in the debate is the one making the claim, not the one testing it.

Mechanism

Why the precautionary case is coherent. It does not require LA to be a toxin. It rests on: a refined industrial food at unprecedented volume (novelty/burden-of-proof), a real oxidation pathway that is worst in heated/fried oils and whose products appear in plaque, a replacement rationale (SFA danger) that has weakened, and randomised trials that lowered cholesterol without lowering deaths. Each is individually modest; together they justify minimising the industrial/fried form rather than trusting it.

Why the toxicity/inflammation claim nonetheless fails. The "omega-6 → arachidonic acid → inflammation" chain is rate-limited in humans, so feeding more LA doesn't flood the inflammatory-eicosanoid pool — which is exactly what the RCTs find (no CRP/IL-6 rise). And tissue-LA biomarker cohorts point the opposite way to the harm narrative. So the honest mechanism is "oxidation-and-novelty concern, especially when fried — not acute human toxicity."

Why "the matrix and the cooking" is the real lever. The clearest harm sits with repeatedly-fried oil and the ultra-processed foods seed oils ride in; the clearest reassurance sits with biomarker status in whole-diet cohorts. Whole-food vs refined, and cold vs fried, is doing most of the legitimate work in both directions.

Risks And Contraindications

• Repeatedly-fried/oxidised oil generates aldehydes/OXLAMs — the one concrete, cross-camp concern; minimise fried and reused oil.
• The "seed oils are toxic" overcorrection can push people toward fried tallow-cooked UPF (feeling virtuous) or toward fearing whole foods that contain LA — both misfires.
• The "eat seed oils freely, it's settled" overcorrection ignores the novelty, COI, weakened-SFA-premise, and failed-benefit trials.
• Don't extrapolate the heated-oil/plaque mechanism to "any LA is toxic," and don't treat the temporal correlation as causal.
• Trans fats are the genuinely-established dietary-fat hazard (de Souza 2015) — not a reason to fear traditional saturated fats.
• Never state or imply that hexane residue causes harm at dietary exposure. The claim this entry makes is that the safety file is old and formally re-opened, and that the alternative is costless — nothing more. The hazard classification and the neuropathy literature describe the bulk solvent and occupational inhalation respectively; presenting either as a food finding is misleading. The single most alarming figure available (the infant ceiling under a maximum-limit scenario) must never appear without EFSA's own finding of no immediate health concern from oral exposure alongside it.
• Never tell a user that seed oils make them burn faster, age their skin, or cause skin cancer. The human causal chain is unsupported, and the one human experiment that moved skin peroxidation hard moved the outcome the opposite way. Do not cite the melanoma-patient tissue-composition datum in any surface — it is a cross-sectional comparison propagated by blog citation with reverse causation unhandled. Equally, do not swing the other way and claim linoleic acid protects skin; the only source pointing that way is a single in-vitro cell paper.
• Do not present the refining contaminants as a present-tense alarm, as a seed-oil-specific charge, or as an adult risk. They are legally capped, they are worst in palm rather than in seed oils, virgin olive oil is exempt because it is not refined, the flagged group was formula-fed infants, and measured levels have declined since the rules came in. And do not say refined oil is "pre-oxidised in the bottle" — the stagewise measurements refute it; the accurate charge is antioxidant depletion and chemical alteration, not oxidation.
• Do not quote pre-mitigation or non-EU contaminant datasets as evidence about oils on a European or UK shelf — the widely-circulated high-exceedance figures trace to a non-EU market not bound by these limits.

Controversy

Nature: a genuinely contested, COI-laden topic with overstatement at both poles.

Position A — "Seed oils are toxic and inflammatory; the cause of modern disease." The skeptic/contrarian take.
• Best evidence: novelty + volume (Blasbalg), documented establishment COI, weakened SFA premise (Siri-Tarino/Chowdhury/de Souza), oxidation chemistry + OXLAMs in plaque, recovered-data RCTs that failed to show benefit.
• Where it overshoots: human RCTs show no inflammatory-marker rise (Johnson & Fritsche); biomarker LA tracks lower mortality (Marklund); the temporal correlation is heavily confounded. "Acutely toxic/inflammatory in humans" is not proven.

Position B — "Seed oils are proven healthy; the science is settled." The establishment/debunker take.
• Best evidence: no inflammation signal, biomarker cohorts favourable, guideline bodies endorse PUFA-for-SFA.
• Where it overshoots: ignores the novelty, the documented COI (P&G/AHA origin; Kearns sugar precedent), the weakened SFA-replacement premise, the oxidation chemistry, and that even Cochrane shows event-reduction without mortality benefit and the recovered RCTs failed.

The funding/bias dimension — cui bono, both ways. The diet-heart consensus seed oils ride on has a documented conflict history (the P&G-sponsored AHA launch; the proven Kearns sugar-industry episode; the criticised 2017 advisory). The alarm side is its own economy (tallow and "seed-oil-free" brands, scanner apps, influencers). The least-conflicted anchors are the establishment-neutral consumption data (Blasbalg), the biomarker cohorts (Marklund), and the recovered-data RCTs (Ramsden) — and together they support precautionary minimisation of the industrial form, no fear of traditional fats, and no proven toxicity.

Realised Position: Minimise refined/industrial seed oils — especially repeatedly-fried — and prefer whole-food and traditional fats; the burden of proof for this recent industrial food, promoted by a conflicted establishment to replace fats demonised on a weakened premise, has not been met. And be honest that the human-toxicity and inflammation claims are unproven, that biomarker linoleic acid tracks lower mortality, and that the temporal correlation is confounded. This is a precautionary, food-form-specific stance, not a toxicology verdict — and it is exactly the recovery-posture default: lean toward the foods we evolved eating when the novel substitute's safety is unsettled.

Cross-Pillar Connections

• Diet (dietary_fat_cholesterol_and_testosterone): the partner entry carrying the omega-6 oxidative-stress/hormonal angle — now consistent with this one (both lean precautionary on refined omega-6, both honest on what's unproven).
• Diet (cholesterol_misinformation_correction): LDL/lipid interpretation; the SFA-reconsideration and "cholesterol fell without mortality benefit" points defer there.
• Diet (processed_unprocessed_red_meat_distinction): the sibling "it's the processing/matrix, and the demonised traditional food was judged on a weak premise" pattern.
• Diet (whole_food_emphasis): the high-value action — whole-food and traditional fats, cut fried UPF.
• Diet (omega3_repletion): ensure omega-3 adequacy rather than fearing the ratio.
• Cross-pillar (cardiovascular_health_management): the hard CV outcomes this evidence bears on.

What would change our mind

Industry bias note

Sources (9)

• Blasbalg TL, et al. (2011), AJCN — US linoleic acid 2.2–2.8% → 7.2% of energy; soybean oil ~1,163-fold rise (1909→1999). (Government-funded, no conflict.) DiNicolantonio JJ, O'Keefe JH (2018), Open Heart — "oxidised linoleic acid hypothesis"; adipose LA 9.1%→21.5% (1959–2008); OXLAMs in plaque.↗
• AHA 2017 Presidential Advisory (Sacks et al., Circulation) + AHA's "Walking Man"/P&G-Duz 1948 origin (worded as sponsorship, per AHA's own rebuttal of a direct-donation claim). Kearns CE, et al. (2016), JAMA Internal Medicine — proven 1960s sugar-industry capture of nutrition science.↗
• Siri-Tarino PW, et al. (2010), AJCN (n=347,747; SFA not significantly associated with CHD); Chowdhury R, et al. (2014), Annals; de Souza RJ, et al. (2015), BMJ (SFA not associated with mortality/CVD; trans fats were). Hooper L, et al. (2020), Cochrane (n=53,758) — SFA reduction cut CV events 17% (best replaced by PUFA), no mortality reduction.↗
• Ramsden CE, et al. (2013), BMJ (Sydney Diet Heart, LA-substitution mortality HR 1.62); (2016), BMJ (Minnesota, n=9,423, cholesterol down 13.8% without mortality benefit). OXLAMs in human plaque: Brooks (1970), Belkner (1991).↗
• Johnson GH & Fritsche K (2012), J Acad Nutr Diet — RCTs: dietary LA does not raise inflammatory markers (the honest constraint). Marklund M, et al. (2019), Circulation (~69,000; higher biomarker LA → lower CVD mortality, HR ~0.78 — the honest constraint).↗
• Refining contaminants: EFSA CONTAM Panel (2016) opinion on 3-MCPD, 2-MCPD and glycidyl esters (7,175 occurrence data points; 3-MCPD BMDL10 0.077 mg/kg bw/day, renal tubular hyperplasia in rats — superseded by the 2018 re-evaluation); EFSA (2018) revised group TDI 2 µg/kg bw/day (BMDL10 0.20 mg/kg bw/day, male rats) — supersedes the 2016 figure. Regulation (EU) 2018/290 (glycidyl ester maxima; recital 5 "glycidol is a genotoxic and carcinogenic compound"; recital 4 on refined vegetable oils). Regulation (EU) 2020/1322 (3-MCPD maxima; recital 3 on the TDI exceedance in formula-only infants; virgin olive oil exempt as unrefined). FEDIOL voluntary 1 mg/kg glycidyl ester commitment (2015, effective September 2017). Health Canada infant-formula surveys 2015 (n=35) and 2019 (n=33) — bound glycidol mean 8.7 → 6.7 ng/g. Malaysian public-funded review (Universiti Kebangsaan Malaysia, FRGS grant) — palm highest 3-MCPD ester content of any vegetable oil, 4–12% diacylglycerol. Stagewise refining oxidation measurements (rapeseed peroxide and p-anisidine values falling crude → refined; tocopherol loss 50–60%; trans isomer rise at 250–270 °C).↗
• Hexane: Directive 2009/32/EC, Annex I Part II — 1 mg/kg in oils and fats; technical-hexane definition. 21 CFR 101.100(a)(3) (incidental additives, basis of the US label exemption), 21 CFR 173.270 (25 ppm, spice oleoresins), 21 CFR 172.340 (5 ppm, fish protein isolate). EU harmonised classification of n-hexane (CAS 110-54-3) — suspected reproductive toxicant Category 2; repeated-exposure classification reported to carry a 5% concentration limit (a labelling cut-off, not a dose–response threshold; unverified against Annex VI). SCF (1996) — oral NOEL 23 mg/kg bw/day, estimated exposure 0.1, margin ~200. EFSA technical report (2024) and call for data (closed December 2025; final risk assessment due 22 November 2027) — additional data needed on identity, impurities, occurrence and toxicology; exposure modelled on regulatory limits, not measured residues; infants under 16 weeks up to 7.8 mg/kg bw/day under the maximum-limit scenario (a ceiling calculation, not comparable to the 1996 actual-exposure margin); no immediate health concern identified for oral exposure. Retail residue survey (40 samples, 23 brands; detected in 36, maximum 42.6 µg/kg, all below the EU limit); method paper (n=4, one sunflower sample at 2.688 mg/kg). (Australia/NZ 0.3 mg/kg, Canada 10 ppm and China 10 mg/kg are reported via a sector journal and are marked reported, not verified. A frequently-cited Korean limit could not be located in any Korean instrument and is not carried here.)↗
• Skin and UV: Takemura N, et al. (2002), Photochem Photobiol 76(6):657-63 — dietary ALA gave a significantly higher skin peroxidisability index yet lower UVB-induced PGE2, with no significant difference in lipid-damage markers (hairless mice). Rhodes LE, et al. (1994), J Invest Dermatol 103(2):151-4 — 10 g/day fish oil, n=15; minimal erythema dose 18.9 → 41.1 mJ/cm² at 6 months and 23.1 ten weeks after stopping; epidermal omega-3 1.8% → 24.2%; irradiated-skin peroxidation markers 6 → 18.5; "oxidizable buffer." Black HS, et al. (1994), NEJM 330(18):1272-5 — n=76, total fat reduction 40% → 21%, actinic keratoses 10 ± 13 vs 3 ± 7, P = 0.001 (does not isolate linoleic acid). Women's Health Initiative dietary-modification trial — n=48,835, 8.1 years, NMSC HR 0.98 (0.92–1.04), melanoma HR 1.04 (0.82–1.32); baseline-fat interaction p = 0.006, higher-baseline melanoma HR 1.48 (1.06–2.07), hypothesis-generating only; NIH/NHLBI-funded (citation handle outstanding — the trial is identified but the specific skin-cancer publication's author, journal, year and volume:page were not recorded; supply them before this figure is quoted as a citation). Black HS & Rhodes LE (2016), J Clin Med 5(2):23 — the authors' own concession on observational data and the peroxidation paradox (does not discuss the Women's Health Initiative). Manosalva C, et al. (2024), Int J Mol Sci 25(19):10385 — linoleic acid reduced UVB-induced ROS and restored glutathione in HaCaT keratinocytes (in vitro; funding statement could not be retrieved — treat as unchecked). Epidermal barrier requirement for linoleic acid as omega-O-acylceramide — foundational dermatology, from the 1930s essential-fatty-acid deficiency work onward.↗
• Funding notation: anchored on establishment-neutral consumption data (Blasbalg), independent SFA-reconsideration and biomarker analyses, and the recovered-data RCTs — none selling an oil. The seed-oil-promoting establishment's COI is documented (P&G/AHA origin worded accurately; Kearns as the proven precedent); the alarm side's tallow/seed-oil-free economy is flagged. On the newer planks the primary anchors are public law and public regulators (EU Directives and Regulations, EFSA, the US CFR, Health Canada) plus publicly- and charitably-funded academic work; competing-solvent sector sources are excluded from any load-bearing figure after two of their claims failed verification. The verdict is precautionary minimisation of the industrial form, not a toxicity claim.*↗

Related entries