Which food preservatives are actually harmful?

Sodium nitrite and sodium nitrate are added to cured and processed meats - bacon, hot dogs, deli meats, sausages - to prevent botulism and maintain the pink color. The concern isn't the nitrite itself but what it becomes: when nitrites encounter high heat and the amino acids in meat, they can form nitrosamines, a class of compounds that are potent carcinogens in animal studies⁴.

In 2015, the International Agency for Research on Cancer (IARC) classified processed meat as a Group 1 carcinogen - meaning there is sufficient evidence that it causes cancer in humans, specifically colorectal cancer². The nitrite-to-nitrosamine pathway is considered one of the key mechanisms, alongside heme iron and heterocyclic amines from cooking.

An important nuance: nitrates also occur naturally in vegetables like spinach, beets, and celery (which is why "uncured" meats using celery powder are somewhat misleading - they still contain nitrates). But vegetable nitrates come packaged with vitamin C and polyphenols that inhibit nitrosamine formation. Processed meats lack these protective compounds, and the high-heat cooking environment (frying bacon, grilling hot dogs) promotes nitrosamine production. The source and context matter enormously.

The NutriNet-Sante cohort study, following over 100,000 adults, found that a 10% increase in ultra-processed food consumption was associated with a 12% increase in overall cancer risk¹. While this implicates many factors beyond just nitrites, processed meats were a significant contributor.

BHA: Officially on the Watch List

Butylated hydroxyanisole (BHA) is an antioxidant preservative used in fats, oils, cereals, snack foods, and food packaging. The US National Toxicology Program lists BHA as "reasonably anticipated to be a human carcinogen" based on sufficient evidence of carcinogenicity in animal studies³. It consistently causes forestomach tumors in rodents.

The FDA still permits BHA in food, partly because humans don't have a forestomach (the rodent organ where tumors formed) and partly because the doses causing tumors in animals far exceed typical human dietary exposure. This is a legitimate scientific argument, but the NTP designation means the concern is not trivial.

BHT (butylated hydroxytoluene), often used alongside BHA, has a more ambiguous profile. Some animal studies actually showed anti-tumor effects, while others showed tumor promotion depending on dose and organ. The overall evidence for BHT carcinogenicity is weaker than for BHA, and some researchers have noted potential protective effects at low doses.

TBHQ, Sodium Benzoate, and Potassium Sorbate

TBHQ (tert-butylhydroquinone) is used in oils, crackers, and fast food. At high doses in animal studies, it has shown stomach tumors and immune effects. The FDA limits it to 0.02% of fat content. It deserves mild caution but the evidence base is thinner than for BHA or nitrites.

Sodium benzoate is widely used in acidic foods and beverages. The main concern is that in the presence of vitamin C (ascorbic acid), sodium benzoate can form benzene, a known carcinogen⁵. This has been documented in some soft drinks at low levels. The FDA tested beverages in 2006 and found most were within acceptable limits, though a few exceeded the drinking water standard for benzene. On its own, sodium benzoate has a long safety record and low toxicity. The practical concern is specific to vitamin C combinations in acidic drinks.

Potassium sorbate is one of the most widely used preservatives, found in cheese, wine, baked goods, and personal care products. It has an excellent safety profile. While very high concentrations have shown genotoxicity in some in vitro studies, dietary exposure levels are far below those thresholds. This is one of the least concerning preservatives in common use.

The evidence quality varies dramatically across preservatives, and this is where many health blogs mislead people by treating all preservatives as equally dangerous.

Strongest evidence of harm: Nitrites in processed meat have the most robust evidence base. The IARC Group 1 classification² is based on epidemiological studies in humans, not just animal models. Multiple large cohort studies consistently show a 15-20% increased risk of colorectal cancer per 50g/day of processed meat consumption. The biological mechanism (nitrosamine formation) is well-characterized. This is not a "maybe" - it's one of the more solid links in nutritional epidemiology.

Moderate evidence: BHA's carcinogenicity in animal models is consistent and reproducible, earning its NTP designation³. The uncertainty lies in cross-species extrapolation. The forestomach tumors in rodents may not be relevant to humans, but BHA also shows some evidence of effects in other organs at high doses. Given that safer antioxidant alternatives exist (tocopherols, rosemary extract), the risk-benefit calculation is unfavorable.

Weaker evidence, often overblown: Many of the preservative fears circulating online cite in vitro studies (cells in a dish) at concentrations far exceeding dietary exposure, or single animal studies that haven't been replicated. Potassium sorbate is a frequent target despite its excellent safety profile. Similarly, citric acid (a natural preservative) sometimes appears on "chemicals to avoid" lists, which is scientifically absurd.

The NutriNet-Sante cohort¹ highlights an important methodological issue: ultra-processed foods contain many additives simultaneously, making it difficult to isolate the effect of any single preservative. People who eat more processed meat also tend to eat fewer vegetables, exercise less, and have other confounding lifestyle factors. Well-designed studies adjust for these confounders, but residual confounding is always possible.

The dose-response relationship matters enormously. Paracelsus's principle - "the dose makes the poison" - applies directly here. BHA at 0.02% of fat content in a serving of crackers is a very different exposure than BHA administered at 2% of total diet to a rat. Responsible risk assessment accounts for realistic human exposure levels, which most in vitro studies do not model.