A Comprehensive Rebuttal to Seed Oil Sophistry

Updated: Nov 6

The popular consciousness has accepted many dietary villains over the course of the last half century, ranging from fat, to protein, to salt, to carbohydrates. More often than not, dietary constituents that have fallen under such scrutiny have been exonerated in time, as more and more scientific data is brought to light. I suspect that there is a growing number of people who are now wrongfully demonizing vegetable oils as well. Both skepticism and generally negative attitudes toward these oils appear to have skyrocketed in recent years, and it can be seen seemingly everywhere.

From what I can tell, almost all of the claims regarding the negative health effects of vegetable oils are essentially rooted in mechanistic research. Mechanistic research includes studies such as cell culture studies, animal studies, in-vitro studies, or even some short term human experiments. Despite the fact that it is absolutely true that this type of research can be incredibly valuable, it is also almost always extremely inappropriate to extrapolate from mechanistic research to population-level health effects. Especially if there is no population-level outcome data that actually agrees with the mechanistic speculation to begin with.

Ultimately, mechanistic studies carry virtually no information about actual human disease risk itself. Keep this in mind as we parse through the lower- and higher-quality evidence as we go along. This will be important as we explore the various claims made about vegetable oils and their interactions with human health. Let’s start with something familiar and dive into cardiovascular disease.

  1. Abbreviations

  2. Cardiovascular Disease

  3. Lipoprotein Oxidation

  4. Heart Disease

  5. Cancer

  6. Total Cancer

  7. Skin Cancer

  8. Adiposity

  9. 2-Arachidonoylglycerol

  10. Energy Intake

  11. Thermogenesis

  12. Type 2 Diabetes Mellitus

  13. Fatty Liver Diseases

  14. Fat Overload Syndrome

  15. Non-alcoholic Fatty Liver Disease

  16. Autoimmune Diseases

  17. Rheumatoid Arthritis

  18. Systemic Lupus Erythematosus

  19. Degenerative Diseases

  20. Age-related Macular Degeneration

  21. Cognitive Decline

  22. Mechanistic Autofellatio

  23. Inflammation

  24. Lipid Peroxidation

  25. Discussion

  26. Bibliography


2-AG, 2-arachidonoylglycerol; 8-oxodG, 8-oxo-7,8-dihydro-20-deoxyguanosine; ACR, acrylamide; AMD, age-related macular degeneration; AMI, acute myocardial infarction; ANA, anti-nuclear antibody; ApoB, apolipoprotein B; BCC, basal cell carcinoma; BCSO, blackcurrant seed oil; CAO, canola oil; CAT, catalase; CB1, cannabinoid receptor 1; CD, conjugated dienes; CRP, C-reactive protein; EE, energy expenditure; EPO, evening primrose; FADS, fatty acid desaturase; FO, fish oil; GLA, gamma-linolenic acid; GPx, glutathione peroxidase; GR, glutathione reductase; HDL-C, high density lipoprotein cholesterol; HUB, healthy-user bias; IVLE, intravenous lipid emulsion; LA, linoleic acid; LAVAT, La Veterans Administration Hospital Study; LBM, lean body mass; LDL, low density lipoprotein; MCE, Minnesota Coronary Experiment; MDA, malondialdehyde; MS, multiple sclerosis; NSAID, nonsteroidal anti-inflammatory drugs; O2H2, hydroperoxides; O6:O3, omega-6/omega-3 ratio; PUFA, polyunsaturated fat; RA, rheumatoid arthritis; RCT, randomized controlled trials; RMR, resting metabolic rate; SCC; squamous cell carcinoma; SDHS, Sydney Diet Heart Study; SLE, systemic lupus erythematosus; SO, soybean oil; SOD, superoxide dismutase; SU, sunflower oil; T2DM, type 2 diabetes mellitus; TC, total cholesterol; TEF, thermic effect of feeding; TFA, trans-fat; TNF, tumor necrosis factor; TPN, total parenteral nutrition; VAT, visceral adipose tissue.



The primary mechanism by which vegetable oils are suggested to increase the risk of cardiovascular disease (CVD) is through the oxidation of polyunsaturated fats (PUFA), particularly linoleic acid (LA), in low density lipoprotein (LDL) phospholipid membranes. In the literature, this hypothesis seems to be spearheaded by DiNicolantonio and O'Keefe (2018) [1]. The argument starts by proposing that saturated fat (SFA) is protective against LDL oxidation because SFA is not vulnerable to the same sort of oxidative damage to which PUFA is vulnerable, and that oxidized LA correlates with some measures of CVD.

It is therefore concluded that limiting one's intake of LA and increasing the intake of SFA protects LDL particles from oxidative damage and thus reduces the risk of CVD. However, does this simplistic mechanistic hypothesis regarding SFA and PUFA actually work this way in real life?

As it turns out, we've thoroughly investigated the effects of altering the fatty acid composition of the diet on LDL oxidation rates in humans. For example, Mata et al. (1996) explored this question with one of the most rigorous studies of its kind. No statistically significant differences between the effects of high-SFA diets and high-PUFA diets on the lag time to LDL oxidation were observed [2].

However, diets high in monounsaturated fat (MUFA) diets might actually be more protective than either SFA-rich or PUFA-rich diets. When comparing LDL that were enriched with PUFA to LDL that were enriched with MUFA, Kratz et al. (2002) observed a stepwise decrease in LDL lag time to oxidation with increasing PUFA, specifically linoleic acid (LA) [3].

Don't get too excited, though. Despite the authors claiming that some studies show that supplemental vitamin E doesn't mask the effects of dietary fatty acids on LDL susceptibility to oxidation, their reference for this claim does not support this conclusion. The work of Reaven et al. (1994) was their supporting reference for this claim and it actually paints a slightly different picture [4]. Lag-time to oxidation was more than doubled with vitamin E supplementation compared to the previous study.

Reaven et al. discovered that the lag time to LDL oxidation was enormously high in all groups after a run-in diet that included 1200mg/day of vitamin E for three months before randomization. Compared to the previous study by Kratz et al. (2002), the LDL lag time to oxidation was 150% higher (~60 minutes to ~150 minutes). This is consistent with other research showing that the vitamin E content of LDL particles linearly increases the lag-time to LDL oxidation [5].

Both studies used the same copper sulfate solution methodology to measure the lag-time to oxidation, as they both reference the same source for the methods developed by Esterbauer et al. (1989) [6]. As such, it wouldn't seem that the increase in LDL lag-time to oxidation seen with vitamin E supplementation would be due to a difference in measurement methods. So we have a decent proof of concept that antioxidants like vitamin E can probably modulate the lag-time to LDL oxidation to an enormous degree.

Which brings me back to the trial by Kratz et al. (2002), which compared olive oil (OO) versus sunflower oil (SU). Despite the fact that all diet groups were receiving roughly the same amount of vitamin E, the OO group ended up with significantly higher vitamin E status compared to the SU group.

There are a few plausible explanations for the results. Firstly, the OO group also started with much higher status to begin with. Secondly, it is well understood that higher PUFA intakes increase vitamin E requirements in humans [7]. Lastly, the run-in diet was high in SFA. Such a diet is almost assuredly to be lower in vitamin E, which could exaggerate the effects of high-PUFA diets.

SFA is not vulnerable to lipid peroxidation in any way similar to that of PUFA. However, SFA being less vulnerable to lipid peroxidation typically means that sources of SFA contain very low amounts of antioxidants [8]. For example, both coconut oil and butter contain negligible vitamin E and have minimal polyphenols. The potential importance of polyphenol antioxidants in protecting LDL from oxidation has been demonstrated in research investigating OO [9].

Figure 2 from Marrugat et al. (2004) shows that virgin OO increases the lag time to oxidation to a greater degree than that of refined OO, with the primary differences between these different OOs being their polyphenol content. Many polyphenols act as antioxidants once they're inside our bodies, so these results are not particularly surprising.

It is likely that the overall dietary pattern matters more than PUFA, or even LA, in altering the susceptibility of LDL to oxidation. This principle has been demonstrated multiple times [10-13].

For example, in a trial by Aronis et al. (2007), diabetic subjects were placed on either a fast food diet that was formulated to be "Mediterranean diet-like" or assigned to follow their usual diet. There was also a third arm of non-diabetics placed on the Mediterranean-style fast food diet. Lag time to LDL oxidation was measured according to the same methodology as described above, with a copper sulfate solution.

It should be noted that the foods were specifically formulated to increase the lag time to LDL oxidation. However, what makes these results more impressive is that the fast food groups (groups A and B) were consuming twice as much PUFA than they were at baseline. Despite this we see some of the longest lag times to LDL oxidation observed in the literature in a population that has not been primed with megadoses of vitamin E for three months.

Altogether this would seem to divulge that diet quality matters more than PUFA, or even LA, for LDL oxidation, because we've seen multiple times that low-PUFA or low-LA diets can be outperformed by high-PUFA or high-LA diets of better overall quality. Little things add up, and the effect of diet is greater than that of MUFA alone, SFA alone, or even polyphenols alone. Perhaps not vitamin E alone, though.

But let's be clear. Measuring the lag time to LDL oxidation in this way is highly contrived, and probably not reflective of normal physiological conditions. When you plop the LDL particles in a copper sulfate solution, eventually all of the LDL will oxidize. There's no escaping it. Naturally, the high-PUFA LDL will oxidize marginally sooner. But ultimately, this does not tell us anything about what would happen to these LDL particles under physiological conditions. For that, it might be more useful to look at markers like oxLDL.

A marker like oxLDL can give us a better sense of just how many oxidized LDL are likely to form in the blood after a particular intervention or in a particular context. This is important because merely looking at the lag time to oxidation could give us an exaggerated sense of what is likely to happen in vivo.

For example, it may be the case that when you expose LDL particles to copper sulfate, they oxidize in under an hour. But under physiological conditions, that same oxidation could potentially take days. If the LDL are cleared from the blood before oxidation can occur, then the results of the copper sulfate test are probably not very informative.

One particularly long crossover-style RCT by Palomäki et al. (2010) compared the effects of butter to that of canola oil (CAO) on oxLDL [14]. Overall the butter diet resulted in higher LDL and higher oxLDL. I wasn't able to find many PUFA-SFA substitution trials that measured oxLDL beyond this one study, unfortunately.

Again, I speculate that this is likely the result of SFA being a poor source of antioxidants. Or perhaps it's because baseline diets could have been higher in PUFA, and reducing vegetable oil intakes might just be cutting off robust sources of antioxidants and increasing LDL oxidation. There are not many studies investigating this, so it's not clear at the moment.

But, just for the sake of argument let's assume that high-PUFA diets do increase LDL oxidation relative to high-SFA diets. Would that actually be a bad thing? Perhaps not. One study by Oörni et al. (1997) has identified that oxidized LDL are less likely to be retained within the arterial intima when compared to native LDL [15]. If the LDL are oxidizing in the serum, perhaps this just opens up more disposal pathways for LDL and lowers its chances of getting retained in the subendothelial space to begin with.

Lastly, oxLDL isn't actually a robust risk factor for CHD. Wu et al. (2006) discovered that the association between oxLDL and CHD does not survive adjustment for traditional risk factors like apolipoprotein (ApoB) or TC/high density lipoprotein cholesterol (HDL-C) [16].

Essentially this means that risk is more closely tracking ApoB or TC/HDL-C, and is not particularly likely to be tracking oxLDL at all. So even if it were the case that diets high in vegetable oils simply increased oxLDL, it wouldn't appear that it moves the needle for risk in the real world. It would also suggest that the hypothetical detriments of increasing LDL oxidation aren't significant enough to offset the LDL-lowering benefits of a high-PUFA diet. As we will discuss later in this section.

In the above study by Wu et al. (2006), the Mercodia 4E6 antibody assay was used to measure oxLDL. Some have argued that this assay is invalid due to supposedly making poor distinctions between native LDL and oxLDL [17]. But, if this was true, wouldn't both models presented in the study above produce similar results? Instead, we see a nearly threefold difference between the results of both models.

It is instead suggested that measures of oxidized phospholipids, such as the E06 antibody assay, are more robust measures of oxLDL than the 4E6 antibody assay. However, the E06 antibody assay is actually more vulnerable to the exact type of confounding that has been suggested for the 4E6 antibody assay. This is explained on Mercodia's website [18].

“The proprietary mouse monoclonal antibody 4E6 is developed by professors Holvoet and Collen at the University of Leuven in Belgium. It is directed against a conformational epitope in the ApoB100 moiety of LDL that is generated as a consequence of substitution of at least 60 lysine residues of Apo B100 with aldehydes (Holvoet 2006). This number of substituted lysines corresponds to the minimal number required for scavenger-mediated uptake of oxidized LDL.”

Essentially, the 4E6 antibody assay makes a clear distinction between native LDL and oxLDL by only binding to ApoB that has been modified sufficiently, as to not be recognizable by LDL receptors. As described on the Mercodia website, the 4E6 antibody only binds to ApoB particles that have >60 of their lysine residues modified by aldehydes, which is the threshold for initiating binding affinity with scavenger receptors, and foreclosing binding affinity with LDL receptors. This makes the 4E6 antibody assay an excellent assay for clearly distinguishing between native LDL and oxLDL.

Mercodia goes on to discuss the reasons for why the E06 antibody assay could be problematic for assessing oxLDL.

“Substituting aldehydes can be produced by peroxidation of lipids of LDL, resulting in the generation of oxidized LDL. However, lipid peroxidation is not required. Indeed, aldehydes that are released by endothelial cells under oxidative stress or by activated platelets may also induce the oxidative modification of Apo B100 in the absence of lipid peroxidation of LDL.”

Because lipid peroxidation of the LDL particle's phospholipid membrane is not required for an LDL particle to oxidize, a measurement of oxPL could easily mistake a minimally oxidized LDL particle as an oxLDL. For this reason, it is likely that the 4E6 antibody assay is likely to better reflect the actual number of oxLDL [19].

This is relevant because the immune cells that mediate the formation of atherosclerotic plaques only tend to take up maximally oxidized LDL particles, not minimally oxidized LDL particles [20][21]. Maximally oxidized LDL have to be disposed of through scavenger receptor-mediated pathways, rather than LDL receptor-mediated pathways.

If minimally oxidized LDL likely contribute as little to foam cell formation as native LDL, why favour measures of minimally oxidized LDL such as the E06 antibody assay over measures of maximally oxidized LDL such as the 4E6 antibody assay?

Unfortunately, so far no studies have attempted to explore the relationship between oxLDL, as measured by the E06 antibody assay, and CHD outcomes after adjustment for total ApoB. The closest we have is a single study by Tsimikas et al. (2006) that found no correlation between oxPL/ApoB and ApoB [22].

However, if ApoB and oxPL tend to vary in tandem, the oxPL/ApoB ratio might not be expected to change very much from subject to subject. If that is the case, then we would not expect oxPL/ApoB to correlate very well at all. It would be nice to see univariable and multivariable models presented that test for independent effects of these biomarkers.

Nevertheless, if merely having more LA in your body meant that you would increase lipoprotein oxidation and thus get more CHD, we would not expect results like those of the meta-analysis conducted by Marklund et al. (2019) [23]. When investigated closely, the higher your tissue representation, the lower your risk of both CHD and ischemic stroke.

It’s important to note that there was not a single statistically significant increase in risk observed in any of the included cohorts when comparing the lowest tissue LA to the highest tissue LA. One of the more interesting findings was from a regression analysis of the Scottish Heart Health Extended Cohort by Wood et al. (1984), which found a strong inverse correlation between adipose LA and CHD incidence [24].

This certainly isn't what we would expect if the chain of causality is LA -> OxLDL -> CHD. However, as we'll discuss in the next section, this is what we'd expect if the chain of causality is SFA -> LDL -> CHD. So, let's go ahead and dive into that data next.


Now that we've established that there is scant evidence validating oxLDL as a robust, independent marker of CVD risk, let's move on to actual human outcomes. It is a common belief among certain diet camps that PUFA actually increases CVD risk, and that SFA may even lower the risk of CVD. But is there evidence for this claim?

Let's start with the epidemiological evidence, being sure to keep the original question in mind— does substituting PUFA for SFA lower CVD rates? There have been multiple meta-analyses that have aimed to answer this question, however the application of meta-analysis methods in this context can produce some significant interpretative challenges. Let me explain why.

The relationship between SFA and CVD is nonlinear, and significantly influenced by the degree of replacement with PUFA. As such, the linear summation of prospective cohort studies can hide the nonlinear effect. For example, let's have a look at one of the most heavily-cited meta-analyses by Siri-Tarino et al. (2010), which investigated the relationship between SFA and CVD in prospective cohort studies [25].

As we can see, the aggregated results are null (P = 0.22). However, the I² (a measure of heterogeneity) is 41%, and no attempt was made to investigate the source of that heterogeneity. If we take the time to calculate the intake ranges of each cohort study (when possible), we can test for nonlinearity.

After stratifying the included studies by intake range, we see a statistically significant 30% increase in risk in subgroup one (~25±15g/day) and null results for subgroups two and three (35±15g/day and 45±15g/day, respectively). This is precisely what we'd expect to see if the risk threshold presupposed by the Dietary Guidelines was correct. McGee et al. (1984) needed to be removed for this analysis because SFA intake ranges could not be determined from the provided data.

But perhaps this is an isolated finding. There are many other meta-analyses on this subject [26-29]. Maybe the other meta-analyses on this subject would find something different? Let's find out.

Applying the same methodology to all of the available meta-analyses yields the same result. This time subgroup two is representing the intake threshold presupposed by the Dietary Guidelines (~25±15g/day). We see similar increases in risk in the same subgroup across all published meta-analyses investigating the relationship between SFA and CVD (RR 1.14 [1.07-1.22], P<0.0001).

You may be asking why subgroups one, three, and four are null, whereas subgroup two is not. The answer is quite simple. It's because those ranges are not crossing the threshold of effect. The range of intake represented in subgroup one exists below the threshold of effect, whereas subgroups three and four exist above the threshold of effect. As such, comparing lower intakes to higher intakes within those ranges does not show an additional increase in risk. The relationship between SFA and CVD is sigmoidal.

Comparing intakes that exist on either the floor or ceiling of the risk curve will typically produce null results. Only when the middle intake range is investigated is the increase in risk observable.

But how does PUFA play into this? To answer this question, let's make our own meta-analyses. Let's take all of the cohort studies that fall within that middle range and plot them out for both SFA and PUFA.

Inclusion Criteria:

  • Prospective cohort studies investigating the relationship between SFA and/or PUFA and CVD.

  • Endpoints directly related to CVD, coronary heart disease, ischemic heart disease, or myocardial infarction (events, mortality, incidences, etc) are acceptable.

  • Risk estimates stratified from lowest to highest SFA and/or PUFA intakes.

  • Cohorts must fit into one of four intake categories:

  1. 0-10g/day (lower range) to 20-30g/day (upper range)

  2. 10-20g/day (lower range) to 30-40g/day (upper range)

  3. 20-30g/day (lower range) to 40-50g/day (upper range)

  4. >30g/day (lower range). Min. 20g/day difference from lower to upper range

Exclusion Criteria:

  • Studies pooling results across multiple cohorts from different countries.

  • Risk adjustments controlling for serum cholesterol and hyperlipidemic medications.

  • Cohorts consisting of people with pre-existing CVD and/or currently taking hyperlipidemic medications.

  • Studies investigating the same cohorts as other included studies. Tie-breakers are decided based on differences in study quality (e.g., chosen subgroups, endpoints, multivariate adjustment models, etc).

Drag-netting the literature yielded a total of 74 studies. 20 studies were excluded due to reporting irrelevant endpoints (e.g., stroke, cerebral hemorrhage, atrial fibrillation, etc). Of the 41 remaining studies, 32 studies were excluded due to including duplicate cohorts, adjusting for blood lipids, being multinational, failing to specify intake ranges, or having an intake range that did not fall into one of the prespecified subgroups.

Results for Saturated Fat:

Altogether there were 21 studies that met all of the inclusion criteria [30-50]. Cohorts were stratified across four subgroups, based on absolute grams-per-day intakes of SFA. Total pooled results across all subgroups showed a non-significant increase in risk (RR 1.07 [0.97-1.18], P=0.16). Subgroup two was the only subgroup to reach statistical significance (RR 1.24 [1.10-1.40], P=0.0005).

Again, we see the exact same thing. The increase in risk occurs in the same subgroup, in the same intake range— the same intake range presupposed by the Dietary Guidelines to increase risk.

Results for Polyunsaturated Fat:

Altogether there were 10 studies that met all of the inclusion criteria. In order to preserve the PUFA to SFA ratio, cohorts were stratified across four subgroups, based on absolute grams-per-day intakes of SFA. Total pooled results across all subgroups showed a statistically significant reduction in CVD risk (RR 0.91 [0.83-1.00], P=0.04). Subgroup two was the only subgroup to reach statistical significance (RR 0.87 [0.80-0.93], P=0.0002).

This second forest plot was not stratified by PUFA intake. As mentioned above, it was instead stratified by SFA intake. This way the ratio of SFA to PUFA would be better preserved. This is important for understanding the relationship between these two dietary fats.

Both SFA and PUFA have inverse sigmoidal relationships with CVD. The more PUFA you replace with SFA, the higher your risk. The more SFA you replace with PUFA, the lower your risk.

This is exactly the same relationship we see in the RCTs as well [51]. Recently, Hooper et al. (2020) published an enormous meta-analysis and meta-regression analysis for the Cochrane Collaboration, which investigated CVD-related outcomes of many PUFA-SFA substitution RCTs.

Their analysis shows that substituting SFA for PUFA, and crossing the 10% of energy threshold as SFA, increases CVD risk in the aggregate. Especially for CVD events, CVD mortality, CHD mortality, and non-fatal acute myocardial infarction (AMI).

Cochrane also performed multiple meta-regression analyses on the included data. Meta-regression is a tool used to investigate the relationship between two variables when they are controlled by a moderator variable. In this case, the two variables in question are SFA intakes (or PUFA-SFA substitution) and CVD. Multiple moderator variables were modeled.

The only statistically significant moderator variable was total cholesterol (TC) (P=0.04). This suggests that the chain of causality is SFA -> TC -> CVD. This therefore suggests that to the extent that a dietary modification reduces TC, reductions in CVD should follow. This is confirmed in their exploratory analysis later on in the paper.

Additional exploratory meta-analyses by Hooper et al. (2020) also further divulge that SFA reduction lowers total CVD events (analysis 1.35), the best replacement for SFA is PUFA from vegetable oils (analysis 1.44), and the effect is likely via lowering TC (analysis 1.51). This evidence dovetails perfectly with the epidemiological evidence discussed above.

Not included in many of Cochrane's analyses were two studies that are often offered up as damning counterevidence, and often cited in support the notion that vegetable oils increase CVD risk. These two trials are the Sydney Diet Heart Study (SDHS) by Woodhill et al. (1978) and the Minnesota Coronary Experiment (MCE) by Frantz et al. (1989) [52-53]. These trials were both designed such that SFA was to be replaced with PUFA in the intervention groups in the form of vegetable oil-based margarines, and PUFA was to be replaced with SFA in the control groups.

Both trials achieved significant differences in TC during the course of both of the trials. However, the SDHS did not achieve a significant difference in the magnitude of the reduction, and the final difference in TC was only around 12mg/dL. While statistically significant, it is questionable whether or not either of these changes in TC were clinically significant. On the other hand, the MCE saw significant reductions in TC.

Neither trial saw a statistically significant difference in either CVD mortality or total mortality.

However, in the aggregate, there is a significant increase in CVD mortality between the two trials. This is concerning, seeing as though these two trials are often touted as the best designed trials that have been conducted in the investigation of this research question, aside from the LA Veterans Administration Hospital Study (LAVAT) by Dayton et al. (1969) [54]. However, the MCE's design ended up allowing participants to enter in and exit out of the trial at their leisure, and this ended up resulting in a mean follow-up time of around 1.5 years. Which is abysmal.

It should be noted that the SDHS was a secondary prevention trial, which means that the subjects had already had a single CVD event at the time of enrollment. Given the questionable clinical relevance of the final differences in TC, it is even more ambiguous how meaningful the findings were. It has been observed that differences in SFA intake don't always change CVD outcomes in high-risk populations [55].

Nevertheless, we're left wondering precisely why we didn't see the predicted effect in either of these trials. Aside from the revolving door protocol that was used in the MCE, the designs were not particularly terrible as far as large-scale RCTs go. So, why didn't we see the effect that we see in nearly every other trial of similar design and quality, as mentioned above?

The answer likely lies in the types of fat-based products that were provided to the vegetable oil groups in these two trials. The vegetable oil groups were the groups that were meant to increase PUFA intake. PUFA was provided to subjects primarily in the form of corn oil (CO) based margarines. This is a problem. Margarines are typically made primarily from unsaturated fats, which are liquid at room temperature. In order to make the product solid and spreadable, the oils need to be hardened. There are essentially two standard ways to achieve this effect.

First, we can harden oils through hydrogenation or through emulsification. However, non-hydrogenated margarines were not commonplace on the market prior to the 1970s [56]. Both the SDHS and the MCE were conducted during the 1960s. At the time, hydrogenation was the preferred method to harden margarines. However, this has the secondary effect of generating trans fatty acids (TFA).

TFAs are associated with an increased risk of many diseases, including CVD. However, TFAs are the only known fatty acid subtypes that have been associated with increased CVD mortality when replacing SFA on a population-level [57].

Basically, TFAs are fucking dangerous— more dangerous than SFA. The CO-based margarine used in the SDHS was "Miracle" brand margarine [58]. According to Dr. Robert Grenfell of the National Cardiovascular Health Director at the Australian Heart Foundation and the Deputy Chairman of the Sydney University Nutrition Research Foundation, Bill Shrapnel, Miracle brand margarine was up to 15% TFA at the time the study was conducted [59].

“When this study began, Miracle margarine contained approximately 15 per cent trans fatty acids, which have the worst effect on heart disease risk of any fat. The adverse effect of the intervention in this study was almost certainly due to the increase in trans fatty acids in the diet”

If Woodhill et al. (1978) truly fed the subjects in the vegetable oil group (group F) as much of that margarine as they seem to be claiming to in their publication, that could be around 5.7g/day of TFA.

The margarine used in the MCE was likely to be Fleischmann's Original, due to the availability and popularity at the time of the intervention. As well as the fact that the product was developed in direct response to pro-PUFA research that was being conducted at the time.

The only other two margarines that were potentially available in the region at the time the MCE was conducted were Imperial margarine and Mazola margarine. Some Imperial products remained high in TFA until very recently [60]. Mazola was pretty much the only non-hydrogenated margarine on the market at the time.

The answer to the question of which margarine was used in the MCE is anybody's guess at this point. But if I were generous, I'd say that based on the margarines that were available (and most likely to be used) at the time the MCE was conducted, there is roughly a 67% chance that the trial was potentially confounded by TFA in the vegetable oil diet. Though Ramsden et al. (2016) remain skeptical that confounding such as this was likely [61].

However, even if the MCE was not confounded by TFA, we would still have a good reason to exclude it from consideration. Like I mentioned earlier, the trial was designed in such a way that the participants had the liberty to enter and exit the trial at their leisure. In the aggregate, there was only a 12-18 month follow-up time. As such, the trial has significantly less power than all other trials that investigated this particular sort of dietary substitution.

Ramsden et al. have argued at length that the control group was likely confounded by TFA to a greater degree than the vegetable oil group. In their view, this could indicate that LA is actually worse for you than TFA. Essentially they claim that vegetable shortening was added to the control group in order to boost the SFA content of the diet, and because vegetable shortening is hydrogenated, it must have contained a large amount of TFA. This is dubious.

Fully hydrogenated oils contain very little TFA [62]. However, partially hydrogenated oils such as most soft margarines of the period contain high levels of TFA. Here is a selection of SOs taken from the USDA's SR28 Nutrient Database [63].

As you can see, a tablespoon of partially hydrogenated SO margarine contains 2200% more TFA than a tablespoon of fully hydrogenated SO shortening.

Ramsden et al. (2016) also acknowledge that the vegetable oil group's diet used soft margarines that were likely to contain some TFA. However, they also under-appreciate the fact that fully hydrogenated vegetable shortening contains pretty much the same amount of TFA as its unadulterated, non-hydrogenated counterpart. In fact, this is precisely why fully hydrogenated shortenings are still on the market despite the TFA ban in most developed countries. It's because those fats never had much TFA to begin with.

“Because the trans fatty acid contents of MCE study diets are not available, one could speculate that the lack of benefit in the intervention group was because of increased consumption of trans fat. Indeed, in addition to liquid corn oil the intervention diet also contained a serum cholesterol lowering soft corn oil polyunsaturated margarine, which likely contained some trans fat. The MCE principal investigator (Ivan Frantz) and co-principal investigator (Ancel Keys), however, were well aware of the cholesterol raising effects of trans fat prior to initiating the MCE.77 Moreover, Frantz and Keys previously devised the diets used in the institutional arm of the National Diet Heart Feasibility Study (NDHS), which achieved the greatest reductions in serum cholesterol of all NDHS study sites.2 Hence, it is highly likely that this experienced MCE team selected products containing as little trans fat as possible to maximize the achieved degree of cholesterol lowering. Perhaps more importantly, it is clear from the MCE grant proposal that common margarines and shortenings (major sources of trans fat) were important components of the baseline hospital diets and the control diet (but not the intervention diet). Thus, confounding by dietary trans fat is an exceedingly unlikely explanation for the lack of benefit of the intervention diet.”

Given that we can soundly infer that the SDHS study was confounded by TFA, and the MCE was likely to be confounded by TFA, the results of these trials start to make a lot more sense. The effect sizes, and the direction of effect, mirror what we see in epidemiological research, as well. When TFA replaces SFA, risk goes up. When PUFA replaces SFA, risk goes down. If it is still to be maintained by skeptics that vegetable oils increase the risk of CVD, rather than lowering it, they must satisfy the burden of proof with extraordinary evidence.

I would surely be remiss if I did not briefly acknowledge one trial that was left out of the analysis by Hooper et al. (2020), which was the Lyon Diet Heart Study [64]. The reason this trial is noteworthy is due to it being the only experimental investigation of the diet-heart hypothesis that actually made an effort to reduce LA alongside SFA. The trial also saw one of the largest reductions in CVD risk ever recorded in this body of literature, with a massive 73% reduction in CVD risk over four years. Skeptics of vegetable oils would like to attribute this effect to the lowering of LA in the diet, but is there any merit to this?

The trial did achieve statistically significant differences in LA intake between groups, with 5.3% of energy as LA in the control group and only 3.6% of energy as LA in the intervention group [65]. However, if we calculate the absolute intakes of LA for both groups using each group’s total daily calories, we discover that the absolute differences in LA intake between groups amounted to about 4.5g/day— equal to approximately 1.5 teaspoons of SO. In the most clinical, scientific vernacular, this would be referred to as sweet fuck all.

So, why did the trial see such a massive reduction in risk? Likely because, as far as multifactorial interventions go, this trial did its best to knock the ball out of the park. Just take a look at the diets.

Firstly, the groups achieved a difference of 10g/day in SFA and the between-group differences in SFA cross the 25g/day threshold that we discussed earlier. Secondly, the intervention group was consuming significantly fewer foods that are positively associated with CVD, such as whole meat, processed meat, and butter. Lastly, the intervention group was consuming significantly higher amounts of fruit and vegetables (considered together), as well as unsaturated fats, which are strongly inversely associated with CVD.

Bottom line is that there is a lot going on in these diets that could be strongly modifying CVD risk. To attribute the massive 73% reduction in CVD risk to 4.5g/day of LA is frankly dubious, and not replicated anywhere else in the literature, despite it being directly tested. Virtually all other trials that were well controlled found the opposite effect of increasing LA. Honestly, based on what is currently known about LA, I would sooner suspect that the lower LA in the intervention group was likely working against them, not for them.

Lastly, the relationship between LA and CVD has been investigated using Mendelian randomization (MR). You can think of MR as being a method of investigation that is halfway between an RCT and an epidemiological study. Essentially, you find people with gene variants that influence a particular exposure (such as higher tissue LA) and compare their outcomes to other people who do not have those gene variants.

The study methods of MR also assume that genes are randomly distributed in the population, and this should significantly mitigate the potential for certain types of confounding like the healthy-user bias (HUB). It is generally accepted that this methodology offers superior internal validity for these sorts of research questions than typical methods used in the field of epidemiology. It's essentially giving you the best idea of the magnitude of the effect, independent of confounding, that is achievable without lifelong RCTs.

Even though MR data doesn't really need to be mentioned, because we have actual RCT data on this subject. However, it's worthwhile to add it in here just to hammer home how utterly consistent this evidence actually is. This time the data is being sourced from the UK Biobank, which is a massive prospective cohort study that also leverages a repository of biological samples from over 500,000 subjects [66].