Is it safe to eat Seed Oils?
What survives when industry funded and observationally confounded evidence is set aside (Third revised edition, September 2026) Thomas Hoeg, Lifestack Note on method This note has been created partly with AI. However every citation here has been verified against the published record. This edition also applies prespecified evidence criteria, set out in Section 4, and applies them symmetrically. Studies are excluded where they were funded by parties with a commercial interest in the answer, and observational studies are set aside because of confounding that cannot be removed by statistical adjustment. Both exclusions cut against the argument of this paper as well as for it, and the studies removed are named, described and attributed rather than quietly dropped. Abstract
Industrial seed oils raised dietary linoleic acid from roughly 2% to roughly 7% of total energy across the twentieth century
This paper focuses on the research relating to linoleic acid, which is the component of seed oils which has been researched. The other components of a bottle of seed oil has received nearly no research attention. Estimates of total energy intake from seed oils land around 21% for the whole bottle (USDA). No randomised trial has tested multi-decade exposure at contemporary intakes against clinical outcomes for linoleic acid, which has been the research focus. This review restricts itself to evidence that is neither funded by parties with a commercial stake nor vulnerable to “healthy user” confounding, and names what that excludes. Under those criteria the independently funded randomised evidence is thin and internally divided. A WHO- and NIHR-funded Cochrane review finds that linoleic acid lowers serum cholesterol without demonstrated reduction in cardiovascular mortality, all-cause mortality, stroke or coronary events, at low to very low certainty; an NIH-funded meta-analysis using different inclusion criteria finds roughly a fifth fewer coronary events. Mendelian randomisation, which is immune to healthy user confounding because genotype is assigned at conception, is inconsistent: one analysis finds no effect of linoleic acid on ischaemic heart disease alongside modest benefit for diabetes and blood lipids, another finds lower coronary risk, and all rely on the same pleiotropic desaturase variants. The affirmative case for cardiovascular benefit therefore rests substantially on observational cohorts and on work part funded by manufacturers of the product, including a widely cited 30 cohort pooled analysis supported by a restricted grant from a major margarine producer. This does not establish that seed oils cause harm: the randomised evidence for harm is confined to trials confounded by industrial trans fat, and independently funded mechanistic trials on hepatic fat and inflammatory markers point away from it. The defensible conclusion is that a dietary intervention was adopted at population scale on a surrogate that was never validated against clinical outcomes, and that the gap remains open after six decades. It may feel intuitive that heavily processed foods such as seed oils are unhealthy, but US food policy is now reversing ahead of the evidence in much the same way as the original endorsement was made. 1. Introduction
Rapeseed (canola), soybean, sunflower, corn, safflower and cottonseed oils are the dominant fat sources in industrialised food systems
They are distinguished from traditional culinary fats by their high content of the omega-6 polyunsaturated fatty acid linoleic acid, and by industrial extraction and refining processes developed for non-food applications before being adapted for consumption. Their promotion for human health followed from a narrow observation: substituting polyunsaturated for saturated fat lowers serum total cholesterol. Cholesterol reduction was treated as a surrogate for clinical benefit, and dietary guidance followed. This article asks whether that surrogate was ever validated for this exposure at this magnitude – and, if not, what the evidence base actually consists of once the studies with a commercial interest in the answer are removed.
2. “Seed oils” is not a single exposure
Before any evidential question can be answered the exposure has to be defined, and most of the public debate fails at this step. Traditional safflower oil is roughly 75% linoleic acid; high-oleic safflower and sunflower cultivars now in wide commercial use are closer to 15%. Canola contains about 20% linoleic acid alongside roughly 10% alpha-linolenic acid, an omega-3. Soybean oil sits near 50%. Plenish, a genetically modified low-linoleic soybean oil on the US market since 2014, is lower again. Processing varies as widely as composition. Cold-pressed oils differ from hexane-extracted, refined, bleached and deodorised oils in tocopherol content, oxidation status and minor constituents. The same oil behaves differently as a salad dressing, as a single-use sauté fat, and as a commercial fryer medium held at temperature for hours. Studies treating all of this as one exposure, including several of the historical trials on which the sceptical case rests, cannot answer a question about any particular oil.
3. Historical context and actual intake
Chemical extraction and refining made seed oils suitable for human consumption during the industrial era, and mid-twentieth-century public health campaigns promoting polyunsaturated over saturated fat drove adoption. Linoleic acid rose from approximately 2.2% to approximately 7.2% of total energy availability in the United States, principally from soybean oil, and NHANES data from 1999 to 2014 show intakes continuing to rise across every stratum of age, sex, education, race and income. Two qualifications matter. Mean adult intakes remain within the 5–10% Acceptable Macronutrient Distribution Range, and the 2010 FAO/WHO expert consultation set an acceptable range for linoleic acid of 2.5–9% of energy. Current consumption is therefore not an overdose by the standing recommendations. What it is, unambiguously, is a change without historical precedent, adopted at population scale, on the strength of a biomarker.
4. Evidence criteria used in this review
Nutrition research is unusually exposed to two failure modes, and this review handles both explicitly rather than case by case. The first is funding. Almost every food has an industry behind it, and in the United States several of those industries fund research through checkoff programmes – mandatory levies on producers, administered by the Department of Agriculture, whose statutory purpose is promotion of the commodity. Soybeans, dairy and beef all operate such programmes. Studies funded by a party with a commercial interest in the outcome are excluded here and named in Section 7. The rule is applied in both directions: it removes work favourable to seed oils and work favourable to saturated fat alike. The second is confounding. People who choose seed oils because they believe them healthy also exercise more, smoke less and attend medical appointments more often. No statistical adjustment removes this, because the adjusting variables are themselves measured with error and the unmeasured ones cannot be adjusted for at all. Observational cohorts are therefore set aside here and named in Section 8. Two qualifications on the method itself, stated so that the reader can discount this review's conclusions appropriately. Funding exclusion is blunt: the honest version of the argument is not that industry funded findings are false, but that they should be down weighted and then checked against independently funded work. That check is run below, and it is the strongest form of the argument in this paper, because it does not depend on the assumption that sponsorship corrupts. Second, excluding all observational evidence is stronger than the confounding problem strictly requires, and applied consistently across medicine it would discard the evidence linking smoking to lung cancer. It is used here as a stress test; what remains when the most vulnerable designs are removed, not as a claim that observational evidence is worthless. 5. The randomised evidence
Several trials tested substitution of polyunsaturated for saturated fat. None tested contemporary intakes over contemporary timescales, and every one carries confounding that was invisible to its investigators.
5.1 Minnesota Coronary Experiment (1968–73)
Recovered data from 9,423 institutionalised participants showed that replacing saturated fat with corn oil lowered serum cholesterol without producing a mortality benefit; in some analyses greater cholesterol reduction was associated with higher mortality, particularly above age 65. The original investigators did not publish these findings. The recovery and reanalysis were carried out by Ramsden and colleagues under the intramural programme of the National Institute on Alcohol Abuse and Alcoholism at the NIH, with no industry funding. The trial's limitations are severe and should be stated rather than left to a critic. Only about a quarter of participants remained enrolled for a year or more, a very short exposure for endpoints driven by decades of atherogenesis. The intervention delivered linoleic acid partly as corn-oil margarine, which in that era almost certainly contained industrial trans fat that was neither measured nor controlled. Autopsy data were partial and LDL subfractions were never measured. Ramsden's own conclusion is correspondingly narrow: the data show cholesterol lowering without demonstrated reduction in coronary or all-cause death. Failure to demonstrate benefit is not demonstration of harm.
5.2 Sydney Diet Heart Study (1966–73)
This secondary-prevention trial in 458 men replaced saturated fat with safflower oil and safflower-oil margarine and found higher all-cause, cardiovascular and coronary mortality in the intervention arm. It is the most frequently cited trial in the case against seed oils and it is also the most compromised, a point the sceptical literature generally omits. Industrial margarine of that period was partially hydrogenated and high in trans fatty acids, whose cardiovascular risk had not yet been identified and which were therefore neither measured nor adjusted for. A further complication is that the control group also modestly increased its polyunsaturated fat intake. Because there is no comparator for the control arm's own change this cannot show that linoleic acid was harmless, but it does mean the contrast between arms was smaller than the headline implies, and that the trans-fat-laden margarine remains the simplest explanation for the difference. With 458 participants, men only, single-blind design and a median follow-up of 39 months, the trial cannot support strong conclusions in either direction. Safflower oil at roughly 75% linoleic acid with essentially no omega-3 is also a poor model of contemporary exposure.
5.3 Los Angeles Veterans Administration Study
This trial substituted a vegetable oil mixture for saturated fat and reported fewer atherosclerotic events in the intervention group. Its recognised weaknesses are a free-living control arm with poorer dietary control, imbalanced baseline smoking and substantial attrition. It is reported here in full because a review that hedges only against trials with inconvenient results is not a review.
5.4 What independent systematic review makes of all this
The Cochrane review of omega-6 fats pooled 19 randomised trials with roughly 6,461 participants, including previously unpublished data. It was requested and funded by the World Health Organization's nutrition guidance expert advisory group with NIHR Cochrane infrastructure support, and its lead author declared no known conflicts of interest. It is the most extensive independent synthesis of this question in existence. Its findings: increasing omega-6 fat lowers total cholesterol but not other lipid fractions or adiposity. There is no evidence that it reduces cardiovascular mortality, all-cause mortality, stroke or coronary events. There is a possible modest reduction in myocardial infarction, with approximately 53 people needing to increase intake to prevent one event. Certainty of evidence is rated low to very low throughout. Set against this is the Mozaffarian, Micha and Wallace meta-analysis of 2010, funded by the National Heart, Lung, and Blood Institute and a Searle Scholar Award, which previous editions of this paper excluded precautinarily on the basis of the lead author's later industry disclosures. That was an error under this paper's own rule, which turns on who funded the study, not on the author's wider career; the paper is publicly funded and is readmitted here. It pooled eight trials with 13,614 participants and 1,042 coronary events and found a 19% reduction in coronary events (relative risk 0.81, 95% confidence interval 0.70 to 0.95) when polyunsaturated fat replaced saturated fat, with larger effects in longer trials. The two syntheses disagree because they include different trials and ask slightly different questions. Mozaffarian pooled trials that raised total polyunsaturated fat, several of which supplied meaningful omega-3 alongside linoleic acid, and included studies the Cochrane team rated at high risk of bias; Cochrane restricted itself to omega-6 specifically, added recovered data from Minnesota and Sydney, and graded certainty formally. Neither is industry-funded. The honest reading is not that one is right, but that the independently funded randomised evidence is sparse enough that its verdict turns on inclusion criteria – which is itself the finding. It does not establish that seed oils cause harm. It does not securely establish the cardiovascular benefit on which their promotion was based either.
6. Mendelian randomisation: the design that answers the confounding objection
Because the healthy user problem is the central objection to observational nutrition research, it is worth noting that a design exists which is structurally immune to it. Mendelian randomisation uses genetic variants as instruments for an exposure. Genotype is assigned at conception, before any lifestyle choice is made, and does not correlate with whether a person exercises, smokes or reads nutrition labels. It cannot be confounded by health-consciousness. Zhao and Schooling applied this to linoleic acid using an ischaemic heart disease dataset of up to 76,014 cases and 264,785 controls. Genetically predicted higher serum linoleic acid was associated with lower diabetes risk (odds ratio 0.97, 95% confidence interval 0.96 to 0.99, per percentage point increase in linoleic acid as a share of total fatty acids) and with lower LDL and total cholesterol – but showed no association with ischaemic heart disease. The findings were robust to different variant selections, analytic methods and correction for multiple testing. The authors describe the pattern as paradoxical and note explicitly that it is consistent with the Sydney Diet Heart Study and the Minnesota Coronary Experiment: linoleic acid moves the risk factors without moving the disease. A related multi-outcome analysis found genetically predicted linoleic acid associated with lower large artery stroke and venous thromboembolism. Its own authors caution that FADS1 and FADS2 variants influence several fatty acids on the same pathway simultaneously and are therefore imperfect instruments, which is the principal limitation of this design in this application. Previous editions of this paper stopped there and presented Mendelian randomisation as independently confirming the Cochrane null. That was selective. Park and colleagues, using CARDIOGRAMplusC4D summary data and an allele score analysis in UK Biobank, found that genetically higher linoleic acid was associated with lower risk of both coronary artery disease and myocardial infarction, with the result holding under pleiotropy robust methods, while arachidonic acid was associated with higher risk. Mazidi and colleagues combined a meta-analysis of trials with a Mendelian randomisation analysis and reached a similarly mixed picture, and further multivariable analyses of omega-6 and cardiovascular mortality have appeared in 2025. All of these studies draw their instruments from the same FADS1 and FADS2 region, so they share the pleiotropy problem rather than resolving it. Mendelian randomisation is therefore the closest thing to causal evidence obtainable without a trial, it is not industry funded and it is not confounded by lifestyle. But on the question that matters it is inconclusive. What it consistently shows is that genetically higher linoleic acid lowers LDL and total cholesterol and lowers diabetes risk. Whether that translates to fewer coronary events depends on which analysis one reads. That is a weaker result than the previous editions claimed, and it removes one of the two “clean lines” on which their conclusion rested. 7. Studies excluded on funding grounds
These studies are widely cited in support of seed oil safety and benefit. Each is named here with what it found and who paid for it, so that readers can weigh them independently rather than take their exclusion on trust.
Study, what it found and who funded it
(Table missing – will be inserted later) Marklund et al., Circulation 2019 – individual-level pooled analysis of 30 prospective cohorts, 13 countries, using circulating and tissue fatty acid biomarkers. Higher linoleic acid associated with lower total cardiovascular disease, lower cardiovascular mortality and lower ischaemic stroke. The most influential single piece of evidence in favour of seed oils. Unilever provided Tufts University with a restricted grant partly supporting the analysis. Two authors declared Unilever research support for this work. The senior author declared personal fees from Bunge, among others. Unilever and Bunge are, respectively, a leading margarine manufacturer and one of the world's largest oilseed processors. Nagra et al., Critical Reviews in Food Science and Nutrition 2026 – scoping narrative review titled “Concerns about the health effects of industrially produced seed oils are without scientific foundation”. Concluded that the totality of clinical and observational evidence supports the safety and health benefits of seed oils. One of the four authors (Messina) is affiliated with the Soy Nutrition Institute Global, funded through the US soy checkoff – a mandatory levy on soybean farmers, administered by the Department of Agriculture, whose statutory purpose is the promotion of soybeans. The other three are academic or clinical affiliations. The Institute issued the press release. Plasma linoleic acid and inflammation, presented at NUTRITION 2025 (~1,894 participants) Higher plasma linoleic acid associated with lower C-reactive protein, glycoprotein acetyls, serum amyloid A, glucose, insulin and HOMA-IR. Widely reported as debunking the pro-inflammatory hypothesis. Reported as funded through the Indiana University Foundation; the ultimate source of the gift is not stated in the abstract. The presenting author also directs a contract research organisation with food-industry clients. No conflict is established, and in any case the analysis is cross-sectional and would be excluded on design grounds under Section 8 regardless of funding. Listed here because it is widely cited as an answer to the funding objection. Mozaffarian et al., PLoS Medicine 2010 – meta-analysis of substitution trials. READMITTED IN THIS EDITION Estimated roughly 19% fewer coronary events when polyunsaturated fat replaced saturated fat. Frequently cited against the Cochrane conclusion. Funded by the National Heart, Lung, and Blood Institute (NIH) and a Searle Scholar Award from the Chicago Community Trust. Publicly funded. Earlier editions excluded it on the strength of the lead author's subsequent industry disclosures; that misapplied this paper's own criterion, which concerns study funding. Now discussed in Section 5.4. The row is retained so the correction is visible. On the soy checkoff review
The 2026 review deserves a sentence of its own. It is titled, without qualification, “Concerns about the health effects of industrially produced seed oils are without scientific foundation.” Its author affiliations include the Soy Nutrition Institute Global, which exists because American soybean farmers are required by federal statute to hand over half a percent of the sale price of every bushel to fund the promotion of soybeans. The conclusion may well be correct – much of it is consistent with the independent evidence reviewed above. It is nonetheless precisely the conclusion one would have predicted from the letterhead, and a paper announcing in its title that the concerns about its sponsor's product are without foundation has made itself easy to set aside.
The pattern worth noticing
Applied honestly, this exclusion should have removed evidence from both sides in the previous edition of this review. It largely did not, and the reason is instructive. The work most critical of seed oils is publicly funded: the Minnesota and Sydney reanalyses came out of the NIH intramural programme, the Cochrane review was funded by WHO and NIHR with no declared conflicts, and even the rodent work most damaging to soybean oil was supported by the National Institute of Environmental Health Sciences and an NIH R01 grant. But the pattern is less clean than the second edition claimed: the meta-analysis most favourable to seed oils, Mozaffarian 2010, is also NIH-funded, and its earlier exclusion had artificially sharpened the contrast. What survives is narrower – that the pooled cohort evidence and the most prominent 2026 review carry industry money, while the randomised syntheses on both sides do not. Symmetry nonetheless requires acknowledging what the rule removes on the other side. Dairy and beef checkoff programmes fund a substantial literature favourable to saturated fat, structured identically to the soy checkoff. The PURE study received unrestricted grants from multiple pharmaceutical companies. A large share of the popular anti-seed-oil literature is produced by authors with commercial interests in supplements, books or tallow-based products. None of that is admitted here either. 8. Studies excluded on design grounds: the observational literature
These are the studies most often cited in public discussion of seed oils, in both directions. They are excluded not because their investigators did anything wrong, but because
the design cannot separate the oil from the person eating it. Someone who cooks with rapeseed oil because a doctor recommended it is a different person, in dozens of measured and unmeasured ways, from someone who does not and those differences, not the oil, may be producing the outcome. The problem is structural and no amount of statistical adjustment resolves it.
Study, what it found and why excluded
(Table missing – will be inserted later) JAMA Internal Medicine 2025 – butter and plant-based oils, ~221,000 people across three Harvard cohorts, 30+ years Highest butter intake associated with 15% higher total mortality; highest plant-oil intake with 16% lower mortality. Substituting 10 g/day of butter with plant oil modelled at roughly 17% lower total mortality. Food-frequency questionnaire cohort. Plant oil use in these cohorts tracks strongly with education, exercise, non-smoking and medical engagement. The largest and longest dataset available, and still unable to isolate the exposure. Not industry-funded. PURE (Dehghan et al., Lancet 2017) – 18 countries High carbohydrate intake (>60% of energy) associated with higher total mortality; total fat and fat subtypes associated with lower total mortality. Total fat and subtypes not associated with cardiovascular disease or myocardial infarction. Observational, and frequently misused in this debate: it does not evaluate seed oils at all. Dominated by low- and middle-income countries where the high-carbohydrate extreme reflects poverty rather than dietary choice. Also received unrestricted pharmaceutical grants. Dietary and biomarker linoleic acid and mortality – Li et al., AJCN 2020 systematic review and meta-analysis Higher linoleic acid intake and status associated with lower mortality. Pooled observational cohorts. Pooling many confounded studies increases precision without removing bias. UK Biobank plasma omega-6 and cancer (>250,000 participants) Both omega-6 and omega-3 status inversely associated with overall cancer incidence. Observational. Included here for completeness because it points against harm, and is excluded on the same grounds as the studies pointing toward benefit. Colorectal cancer dose-response meta-analysis, Nutrition & Diabetes 2025 Roughly 15% higher colorectal and 30% higher rectal cancer risk comparing extreme categories of linoleic acid intake. Observational, and excluded despite pointing toward harm. The second edition wrongly described a “separate” 20-cohort analysis of 787,490 participants finding no association; that is the same paper. Within it, linoleic acid specifically showed the positive association while total omega-6 and arachidonic acid did not. A within-study inconsistency that itself argues for caution. NHANES analyses of polyunsaturated fat and mortality Associations vary substantially by polyunsaturated fatty acid subclass and by population. Observational, with dietary recall error compounding the confounding problem. Sadeghi et al., Journal of Translational Medicine 2025 – umbrella review and updated meta-analysis of 150 cohorts Higher dietary and circulating omega-6 associated with lower cardiovascular disease, cancer incidence and all-cause mortality overall, with higher risk for some cancers and no protective association in people with pre-existing disease. Observational. The largest current pooled analysis, and the one most likely to be cited against this paper; included so that its exposition is explicit rather than silent. Note the symmetry. Excluding observational evidence removes the strongest studies pointing toward benefit and the strongest pointing toward harm. Anyone who invokes the healthy-user objection to dismiss the JAMA Internal Medicine substantiation analysis is obliged to apply it to the colorectal cancer meta-analysis as well. 9. Mechanistic and animal evidence
9.1 Rodent models
Deol and colleagues found that mice fed a 40% kcal-fat diet in which the fat was soybean oil showed greater weight gain, adiposity, glucose intolerance and insulin resistance than mice fed the same fat load as coconut oil, with hepatic accumulation of polyunsaturated fatty acid metabolites and dysregulation of cytochrome P450 genes. The work was funded by an NIEHS training grant, an NIH R01 and a UC Riverside seed grant, with no industry support. Follow-up work implicated omega-6 and omega-3 oxylipins as mediators and found that Plenish, a low-linoleic soybean oil, produced less obesity, evidence pointing at linoleic acid specifically rather than soybean oil generally. Further work from the same group, published in the Journal of Lipid Research in late 2025, found that transgenic mice expressing an alternative form of the liver regulator HNF4α produced far fewer linoleic-acid derived oxylipins and were largely protected from soybean-oil induced obesity on the same diet, with only hepatic (not circulating) oxylipin levels tracking body weight. The senior author's own summary was that the culprit is not the oil or linoleic acid as such but what the fat is converted into, which shifts the hypothesis from intake toward individual variation in metabolism.
This is a coherent, independently funded and unusually well developed mechanistic programme, and its limits should be stated as plainly as its findings. The comparator was coconut oil at a fixed high fat load, so the result is a ranking between two fats under obesogenic conditions rather than a statement about habitual human intake. Mice differ from humans in desaturase activity, linoleic acid handling and thermoregulatory energetics, and rodent obesogenicity has translated poorly to human outcomes across many exposures. Nothing here licences a claim about people eating seed oils at 7% of energy. 9.2 Hepatic fat: where controlled human trials contradict the rodent story
The rodent work predicts that high linoleic acid promotes hepatic steatosis. Randomised human feeding trials find the opposite. In the LIPOGAIN overfeeding trial, participants overfed muffins made with sunflower oil or palm oil for seven weeks accumulated liver fat on the saturated fat arm, not the polyunsaturated one. A later randomised overfeeding trial found saturated fat increased liver fat by roughly 50% in relative terms, along with liver enzymes and atherogenic serum lipids, with no comparable effect from omega-6 polyunsaturated fat. Bjermo and colleagues found n-6 polyunsaturated fat reduced liver fat relative to saturated fat in abdominal obesity. These are randomised trials, they survive the design filter, and their funders could not be fully verified – a limitation noted here rather than glossed. This is the clearest instance in the whole debate where a mechanistic hypothesis has been tested directly in humans under controlled conditions and failed. Any argument that high linoleic acid drives fatty liver disease in humans has to account for them.
9.3 Inflammation
The pro-inflammatory hypothesis holds that dietary linoleic acid is converted to arachidonic acid and thence to inflammatory eicosanoids. Human conversion is in fact minimal and tightly regulated, and a systematic review of randomised controlled trials in healthy persons found that increasing dietary linoleic acid did not raise inflammatory markers. That review is randomised evidence and survives both filters. The premise underpinning much of the popular case against seed oils is therefore not supported by the evidence this paper is prepared to admit.
A different mechanism deserves more attention than the inflammation claim has received. In a randomised, double-blind, 12-week feeding trial published in June 2026, Sergeant and colleagues supplied 52 healthy adults with at least 95% of their dietary fat as linoleic acid, contrasting a low-linoleic diet at 2.5% of energy with a high-linoleic diet at 10% of energy, with alpha-linolenic acid held equal in both arms. The high-linoleic diet significantly lowered plasma eicosapentaenoic acid and related long-chain omega-3 fatty acids and altered lipoxygenase-derived oxylipins. This is a small, short, publicly registered trial with biomarker rather than clinical endpoints, but it is precisely the kind of controlled human comparison Section 12 calls for, it survives both of this paper's filters, and it supports a mechanism – omega-3 displacement rather than inflammation – that neither camp has foregrounded.
9.4 Thermal degradation: the better-evidenced concern
A separate claim, frequently conflated with the above, holds up considerably better. Linoleic acid is a substrate for lipid peroxidation, and heating linoleic-acid-rich oils to frying temperature generates 4-hydroxynonenal, acrolein, acetaldehyde, formaldehyde and related aldehydes which are incorporated into the fried food. Measurable 4-HNE has been documented in French fries from fast-food restaurants, and concentrations rise with heating time and oil reuse. These compounds are cytotoxic and mutagenic. This is a claim about cooking practice, commercial fryer management and food processing rather than about linoleic acid as a nutrient. It applies with greatest force to commercial deep-frying and comparatively little to cold or low-heat use of fresh oil. It is the most defensible concern in this entire literature and it deserves separation from the general indictment, which has crowded it out.
9.5 Cancer: genuninely unsettled
A 2025 study in Science identified a mechanism by which linoleic acid binds fatty acid binding protein 5 and promotes growth in triple-negative breast cancer, a subtype in which that protein is abundant. This is a mechanistic finding in one tumour biology, not a population dietary claim, and it has been widely over-generalised. The population evidence is observational and excluded above in both directions. No trial has shown that reducing linoleic acid improves cancer outcomes. This is where uncertainty is most defensible and further work most warranted.
9.6 Mitochondria, oxidised lipids and the long-term accumulation argument
The previous sections engage with the case against seed oils as it is usually put in the cardiovascular literature: inflammation, cholesterol and coronary events. That is not the case the most prominent critics actually make. Catherine Shanahan, whose book Dark Calories (2024) popularised the term “hateful eight” for canola, corn, cottonseed, grapeseed, rice bran, safflower, soybean and sunflower oils, argues something different: that linoleic acid accumulates in tissue over years, that its peroxidation products progressive deplete antioxidant defenses and damage mitochondrial membranes, and that the result is a reduced capacity to oxidise fat and generate ATP, forcing cells toward glycolysis and producing insulin resistance. She published this formally as “The energy model of insulin resistance” in Frontiers in Nutrition in April 2025. Because the concern is cumulative and slow, it deserves a separate treatment from the acute claims above, and a separate assessment of what the available studies can and cannot detect.
Three components of the argument are well supported. First, accumulation is real. Linoleic acid cannot be synthesised by the body, is stored in adipose tissue, and has a tissue half-life of roughly two years, closely tracking dietary intake. A person who has eaten a modern diet for twenty or thirty years therefore carries a tissue composition that reflects that history and that would take one to two years to change. Second, the oxidised products are real and diet-responsive. Oxidised linoleic acid metabolites are elevated in non-alcoholic steatohepatitis and Alzheimer's disease, and a 12-week linoleic-acid-lowering diet in an NIH trial significantly reduced circulating oxidised linoleic acid metabolites and the linoleic acid content of plasma lipid fractions. Third, thermal degradation products such as 4-hydroxynonenal are cytotoxic, as Section 9.4 already sets out.
The mitochondrial step is where the argument runs into contrary evidence. Linoleic acid is the major fatty acid of mitochondrial membranes and makes up nearly 90% of the acyl chains of cardiolipin, the phospholipid that binds and stabilises the respiratory chain enzymes; cardiolipin linoleic acid content correlates positively with cytochrome c oxidase activity in rat muscle. In a 2018 Journal of Biological Chemistry study, when docosahexaenoic acid displaced linoleic acid from cardiac cardiolipin the activities of respiratory complexes I, IV and V fell, and reintroducing linoleic acid to the isolated mitochondria rescued both the phospholipid composition and the enzyme activities. This is rodent and in-vitro work and it shows that linoleic acid is required, not that more is better; but a model in which linoleic acid is a mitochondrial poison has to explain why the electron transport chain is built on it.
Controlled human feeding evidence on the metabolic endpoints Shanahan emphasises points the other way. A meta-analysis of 102 randomised controlled feeding trials in 4,220 adults found that replacing carbohydrate, saturated fat or monounsaturated fat with polyunsaturated fat produced the most consistent improvements in glycaemia, insulin resistance and insulin secretion. The Uppsala overfeeding trials in Section 9.2 found less liver fat on polyunsaturated than saturated fat. A 2023 NIH trial that lowered dietary linoleic acid alone in women with overweight did not lower arachidonic acid or endocannabinoids, a negative result for one specific pathway in her model. Against this, the June 2026 Sergeant trial in Section 9.3 found that a high-linoleic diet suppressed long-chain omega-3 status, which is one plausible route to the membrane changes she describes, and a small 1990s Finnish crossover study reported higher urinary F2-isoprostanes on a high-linoleic diet. There is no human study, in either direction, that has measured mitochondrial respiration, fat-oxidation capacity or ATP production after sustained high linoleic acid intake.
There is no human study, in either direction, that has measured mitochondrial respiration, fat-oxidation capacity or ATP production after sustained high linoleic acid intake.
The durations of these studies are the central problem for anyone whose concern is decades of exposure, and they should be stated plainly. The feeding trials pooled in the
102-trial meta-analysis lasted from days to a few months
. The Uppsala trials ran seven, eight and ten weeks. The Johnson and Fritsche inflammation review pooled trials of similar length. The linoleic-acid-lowering and oxidised-metabolite trials ran twelve weeks, as did the Sergeant trial. Among the outcome trials, only about a quarter of Minnesota Coronary Experiment participants were enrolled for a year or more, the Sydney Diet Heart Study had a median follow-up of 39 months, and the Los Angeles Veterans study ran up to eight years; the trials in the Cochrane review ranged from one to eight years. The rodent work runs for a large fraction of a mouse lifespan (the 2015 Deol study fed mice for 35 weeks) but at a fat load and with a comparator that limit translation. If linoleic acid harms mitochondria through slow accumulation and cumulative oxidative damage, an effect that emerges after twenty or thirty years would not be expected to appear in a twelve-week feeding study of insulin sensitivity, and its absence from such studies is not evidence against it. Equally, the accumulation of linoleic acid in adipose tissue in adipose tissue is not itself evidence of harm; it is evidence of exposure. The honest position is that the short-term controlled data are reassuring on the endpoints they measure, that the biochemistry gives the long-term hypothesis some plausibility, and that no study of the appropriate length and design has been done.
The observational cohorts that do span decades (Section 8) point, if anything, toward benefit, but they carry the healthy user problem this paper has set aside. The Frontiers exchange is worth recording because it is the one place the model has been argued in the peer-reviewed literature. A commentary by Lôpez-Moreno (August 2025) objected that the paper supplies no human studies showing that replacing animal fats with redefined seed oils raises cellular oxidative stress, that its opening figure infers causation from a temporal correlation between oil consumption and diabetes prevalence, that its citation of a large randomised trial showing increased cancer appears to refer to the Minnesota Coronary Experiment which did not assess cancer, and that the paper is a variant of the carbohydrate-insulin model. Shanahan's response (October 2025) opens by conceding that the hypothesis has not yet been directly tested and should be, which is the material point for this review. Beyond that, her rebuttal is stronger on some counts than others. She is right that the commentary did not engage with the adipose tissue accumulation data, which is her best founded premise, and right that her model is not the carbohydrate-insulin model: it concerns redox damage to oxidative capacity, not dietary carbohydrate, and the commentary's several paragraphs on the carbohydrate-insulin model are beside the point. She is on weaker ground in citing the Minnesota and Sydney trials as human evidence for a mitochondrial mechanism: neither measured any mitochondrial or oxidative endpoint, both are confounded by trans fat as Sections 5.1 and 5.2 set out, and mortality in a trans-fat-laden margarine arm does not test her model. Her closing suggestion that the commentator's affiliation with a plant-based diet research group explains his conclusions is an argument this paper cannot wholly disown, since it down weights work by affiliation itself. But an academic interest is not a commercial stake, and the objection does not answer the substantive points. The response does not identify the "human mechanistic studies linking seed oils to altered mitochondrial function" it says the original paper contains. Readers should check the original paper's citations for themselves. On the specific question a long-term consumer might ask, whether decades of intake have impaired the capacity to burn fat, the available answer in 2026 is that
nobody has measured it
. What can be said is that the accumulation is real and reversible over one to two years; that the short-term controlled evidence on insulin sensitivity, liver fat and inflammatory markers favours polyunsaturated over saturated fat; that the mechanisms with the best support are omega-3 displacement and oxidised-lipid load rather than a direct block on ATP synthesis; and that individual variation in desaturase genotype and oxylipin producing enzymes, which the 2025 rodent work highlights, may matter more than population averages. Shanahan's model is a testable hypothesis that has not been tested, which is a different thing from either a demonstrated harm or a refuted one. 10. The confounding nobody can resolve
Linoleic acid reaches people predominantly inside packaged snacks, baked goods, fried foods, sauces and restaurant meals – foods that also differ in refined carbohydrate content, sodium, additives, energy density and satiety. This cuts every way at once. Cohort studies associating higher linoleic acid with better outcomes may be measuring people who cook at home rather than eating fried food. Studies associating seed oils with metabolic disease may be measuring the ultra processed matrix rather than the fat. No observational design separates them and no trial has attempted to. Anyone claiming confident causal knowledge in either direction is claiming more than the data permit.
11. Regulation, processing and the 2015–26 policy reversal
Regulatory attention has focused on specific contaminants rather than on the oils as a class. Eruçic acid in rapeseed prompted restrictions and cultivar development, and modern canola is bred to low levels that EFSA has assessed as safe. That is a solved problem and should not be presented as an open concern. Refining is a more legitimate open question. Hexane extraction, high-temperature deodorisation, bleaching and the resulting formation of oxidation products and processing contaminants have received far less systematic evaluation than the fatty acid profile itself. That absence of data is a genuine gap, distinct from the linoleic acid question.
The policy environment changed sharply in 2025 and 2026. Executive Order 14212 established the Make America Healthy Again Commission in February 2025 and its assessment appeared that May. The 2025–2030 Dietary Guidelines for Americans elevated butter, beef tallow and olive oil while declining to mention seed oils, and in January 2026 the Secretary of Health and Human Services was widely reported to have announced plans to phase fourteen seed oils out of school lunch programmes; the primary policy document should be consulted as this proceeds. Both the reversal and the industry-affiliated rebuttals to it are running ahead of the evidence. The trials that would adjudicate between them have not been done.
12. What would resolve this
A multi-decade randomised trial of habitual linoleic acid intake against hard endpoints is almost certainly not feasible. Several tractable designs would nonetheless narrow the uncertainty substantially, and all of them could be publicly funded. Medium-duration randomised feeding trials of three to five years, powered for metabolic and hepatic endpoints and comparing matched diets differing only in fat source, would test the mechanistic claims directly. Larger and better instrumented Mendelian randomisation studies, using variants that separate linoleic acid from the rest of the desaturase pathway, would resolve the paradox identified in Section 6. Trials holding total energy and food matrix constant while varying oil type would begin to separate the fat from the ultra processed context. Randomised comparison of fresh versus repeatedly heated oils with direct aldehyde exposure measurement would test the most likely of the proposed harms. A feeding trial of at least one to two years, long enough for adipose tissue linoleic acid to turn over, with muscle biopsy mitochondrial respiratory and fat-oxidation measurements as endpoints would be the first direct test of the energy-metabolism hypothesis in Section 9.6. And exposure should be disaggregated by oil, cultivar, processing method and cooking application rather than pooled under a single label.
A feeding trial of at least one to two years, long enough for adipose tissue linoleic acid to turn over, with muscle biopsy mitochondrial respiratory and fat-oxidation measurements as endpoints would be the first direct test of the energy-metabolism hypothesis in Section 9.6. And exposure should be disaggregated by oil, cultivar, processing method and cooking application rather than pooled under a single label.
13. Conclusion
Independently funded randomised evidence establishes that replacing saturated fat with linoleic acid lowers serum cholesterol.
It does not establish that this reduces cardiovascular events, cardiovascular death or all-cause death
, and the Cochrane review – funded by WHO and NIHR, with no declared conflicts, rates the certainty of that evidence low to very low. A second, NIH-funded meta-analysis with broader inclusion criteria finds a reduction in coronary events; the two independent syntheses disagree and the disagreement turns on trial selection. Mendelian randomisation, which cannot be confounded by the health-consciousness of the people eating the oil, consistently finds benefit for diabetes risk and blood lipids and inconsistent findings for coronary events. The mechanistic case for harm rests on a plausible but as yet untested long-term hypothesis about oxidative damage and mitochondrial function, and on better-evidenced concerns about thermal degradation products such as 4-HNE. The acute case for benefit rests on shorter-term trials showing improved lipids, liver fat and glycaemic control, and on observational cohorts that this paper has set aside on confounding grounds.
It does not establish that seed oils are harmful or harmless. The randomised evidence for harm is confined to trials confounded by industrial trans fat, in one case with a control group that raised polyunsaturated intake without harm. Independently funded randomised trials on hepatic fat and inflammatory markers point away from harm. The mitochondrial hypothesis advanced by Shanahan and adopted by some critics is biologically plausible, and the adipose tissue accumulation data are real, but the hypothesis has not been tested in humans and mechanistic human evidence on the endpoints it concerns is absent. The observational cohort evidence – set aside here because it cannot disentangle oil choice from the kind of person who makes it – points, if anything, toward benefit.
References
Bjermo H, Iggman D, Kullberg J, et al. Effects of n-6 PUFA compared with SFAs on liver fat, lipoproteins, and inflammation in abdominal obesity: a randomised controlled trial. American Journal of Clinical Nutrition. 2012;95(5):1003-1012. Csallany AS, Han I, Shoeman DW, Chen C, Yuan J. 4-Hydroxynonenal (HNE), a toxic aldehyde in French fries from fast food restaurants. Journal of the American Oil Chemists' Society. 2015;92:1413-1419. Courville AB, Majchrzak-Hong S, Yang S, et al. Dietary linoleic acid lowering alone does not lower arachidonic acid or endocannabinoids among women with overweight and obesity: a randomised, controlled trial. Lipids. 2023;58(6):271-284. Dehghan M, Mente A, Zhang X, et al. Associations of fats and carbohydrate intake with cardiovascular disease and mortality in 18 countries from five continents (PURE): a prospective cohort study. The Lancet. 2017;390(10107):2050-2062. Deol P, Evans JR, Dhahbi J, et al. Soybean oil is more obesogenic and diabetogenic than coconut oil and fructose in mouse: potential role for the liver. PLOS ONE. 2015;10(7):e0132672. Funded by NIEHS T32 5T32-ES018827, NIH R01DK053892 and a UC Riverside collaborative seed grant. Deol P, et al.; Sladek FM (senior author). P2-HNF4α alters linoleic acid metabolism and mitigates soybean oil-induced obesity: role for oxylipins. Journal of Lipid Research. 2025:100932. doi:10.1016/j.jlr.2025.100932. European Food Safety Authority, CONTAM Panel. Erucic acid in feed and food. EFSA Journal. 2016;14(11):4593. FAO/WHO. Fats and fatty acids in human nutrition: report of an expert consultation. FAO Food and Nutrition Paper 91. Rome; 2010. Fajardo VA, McMeekin L, Saint C, LeBlanc PJ. Cardiolipin linoleic acid content and mitochondrial cytochrome c oxidase activity are associated in rat skeletal muscle. Chemistry and Physics of Lipids. 2015;187:50-55. Guyenet SJ, Carlson SE. Increase in adipose tissue linoleic acid in US adults in the last half century. Advances in Nutrition. 2015;6(6):660-664. Hooper L, Al-Khudairy L, Abdelhamid AS, et al. Omega-6 fats for the primary and secondary prevention of cardiovascular disease. Cochrane Database of Systematic Reviews. 2018;11:CD011094. Requested and funded by the WHO Nutrition Guidance Expert Advisory Group with NIHR Cochrane infrastructure support; lead author declared no known conflicts. Imamura F, Micha R, Wu JHY, et al. Effects of saturated fat, polyunsaturated fat, monounsaturated fat, and carbohydrate on glucose-insulin homeostasis: a systematic review and meta-analysis of randomised controlled feeding trials. PLOS Medicine. 2016;13(7):e1002087. Johnson GH, Fritsche K. Effect of dietary linoleic acid on markers of inflammation in healthy persons: a systematic review of randomised controlled trials. Journal of the Academy of Nutrition and Dietetics. 2011;112(7):1029-1041. Mazidi M, Mikhaiilidis DP, Banach M. Omega-6 fatty acids and the risk of cardiovascular disease: insights from a systematic review and meta-analysis of randomised controlled trials and a Mendelian randomisation study. Archives of Medical Science. 2021. Marklund M, Wu JHY, Imamura F, et al. Biomarkers of dietary omega-6 fatty acids and incident cardiovascular disease and mortality: an individual-level pooled analysis of 30 cohort studies. Circulation. 2019;139(21):2422-2436. Partly supported by a restricted Unilever grant to Tufts; two authors declared Unilever research support. Mozaffarian D, Micha R, Wallace S. Effects on coronary heart disease of increasing polyunsaturated fat in place of saturated fat: a systematic review and meta-analysis of randomised controlled trials. PLOS Medicine. 2010;7(3):e1000252. Funded by NHLBI/NIH and a Searle Scholar Award. Nagara M, Goldmann DM, Belury MA, Messina M. Concerns about the health effects of industrially produced seed oils are without scientific foundation: a scoping narrative review of the clinical and observational evidence. Critical Reviews in Food Science and Nutrition. Published online 28 April 2026. doi:10.1080/10408398.2026.2657527. One author affiliated with Soy Nutrition Institute Global, funded via the US soy checkoff. Park S, Lee S, Kim Y, et al. Causal effects of serum levels of n-3 or n-6 polyunsaturated fatty acids on coronary artery disease: Mendelian randomisation study. Nutrients. 2021;13(5):1490. Ramsden CE, Zamora D, Majchrzsak-Hong S, et al. Use of dietary linoleic acid for secondary prevention of coronary heart disease and death: evaluation of recovered data from the Sydney Diet Heart Study and updated meta-analysis. BMJ. 2013;346:e8707. NIH intramural funding (NIAAA). Ramsden CE, Ringel A, Feldstein AE, et al. Lowering dietary linoleic acid reduces bioactive oxidised linoleic acid metabolites in humans. Prostaglandins, Leukotrienes and Essential Fatty Acids. 2012;87(4-5):135-141. NIH intramural funding. Ramsden CE, Zamora D, Majchrzsak-Hong S, et al. Re-evaluation of the traditional diet-heart hypothesis: analysis of recovered data from the Minnesota Coronary Experiment (1968–73). BMJ. 2016;353:i1246. NIH intramural funding (NIAAA). Rosqvist F, Iggman D, Kullberg J, et al. Overfeeding polyunsaturated and saturated fat causes distinct effects on liver and visceral fat accumulation in humans. Diabetologia. 2014;63(7):2356-2368. Rosqvist F, Kullberg J, Ståhlman M, et al. Overeating saturated fat promotes fatty liver and ceramides compared with polyunsaturated fat: a randomised trial. Journal of Clinical Endocrinology and Metabolism. 2019;104(12):6207-6219. Sadeghi R, Norouzzadeh M, HasanRashedi M, et al. Dietary and circulating omega-6 fatty acids and their impact on cardiovascular disease, cancer incidence and mortality: a global meta-analysis of 150 cohorts and meta-regression. Journal of Translational Medicine. 2025;23:314. [Excluded on design.] Sergeant S, et al. Effect of dietary linoleic acid intake on eicosapentaenoic acid status and lipoxygenase-mediated oxylipin biosynthesis in healthy adults: a randomised controlled trial. Nutrients. 2026;18(11):1814. ClinicalTrials.gov NCT02962128. Shanahan C. The energy model of insulin resistance: a unifying theory linking seed oils to metabolic disease and cancer. Frontiers in Nutrition. 2025;12:1532961. Shanahan C. Response: Commentary: The energy model of insulin resistance. Frontiers in Nutrition. 2025;12:1690496. Author declares ownership of Rebel Well LLC. Shanahan C. Dark Calories: How Vegetable Oils Destroy Our Health and How We Can Get It Back. Hachette; 2024. Sullivan EM, Pennington ER, Sparagna GC, et al. Docosahexaenoic acid lowers cardiac mitochondrial enzyme activity by replacing linoleic acid in the phospholipidome. Journal of Biological Chemistry. 2018;293(2):466-483. Raatz SK, Conrad Z, Jahns L, Belury MA, Picklo MJ. Trends in linoleic acid intake in the United States adult population: NHANES 1999-2014. Prostaglandins, Leukotrienes and Essential Fatty Acids. 2018;133. World Health Organization. Saturated fatty acid and trans-fatty acid intake for adults and children: WHO guideline. Geneva; 2023. Yuan S, et al. Plasma phospholipid fatty acids, FADS1 and risk of 15 cardiovascular diseases: a Mendelian randomisation study. Nutrients. 2019;11(12):3001. Zhao JV, Schooling CM. Effect of linoleic acid on ischaemic heart disease and its risk factors: a Mendelian randomisation study. BMC Medicine. 2019;17:61. Zhang Y, et al. Butter and plant-based oils intake and mortality. JAMA Internal Medicine. Published online 6 March 2025. Harvard NHS/NHS II/HPFS cohorts. [Excluded on design.] Li J, Guasch-Ferré M, Li Y, Hu FB. Dietary intake and biomarkers of linoleic acid and mortality. Systematic review and meta-analysis of prospective cohort studies. American Journal of Clinical Nutrition. 2020;112(1):150-167. [Excluded on design.] López-Moreno M. Commentary: The energy model of insulin resistance: a unifying theory linking seed oils to metabolic disease and cancer. Frontiers in Nutrition. 2025;12:1622160. Dietary intake and tissue biomarkers of omega-6 fatty acids and risk of colorectal cancer in adults: a systematic review and dose-response meta-analysis of prospective cohort studies. Nutrition & Diabetes. 2025. PMID 40251189. Twenty cohorts, 787,490 participants. [Excluded on design.]