Learn → How Food Actually Works → Module 08

The ultra-processed food problem

Ultra-processed food is a precise category — defined by industrial ingredients, not nutrient profile — and the strongest single dietary lever on chronic disease.

10 min read

The ultra-processed food problem

TL;DR. Ultra-processed food (UPF) is not a slur — it's a precise category, NOVA Group 4, defined by Carlos Monteiro in 2009 by an operational test: industrial ingredients no home cook owns. The clinical evidence is Kevin Hall's 2019 NIH metabolic-ward trial: nutrient-matched UPF diets drove 500 extra calories per day and a pound a week of weight gain. Roughly 250 epidemiology papers show dose-dependent harm — every 10% rise in UPF raises disease risks 10–25%, independent of obesity. UPF works through four engineering levers: a destroyed food matrix (soft, dry, calorie-dense); hyperpalatable salt-sugar-fat synergy; mismatched taste signals from non-nutritive sweeteners; and microbiome damage from emulsifiers. Reformulation cannot fix this because the category is defined by ingredients, not macros — "lower sugar Cheerios" is still NOVA 4. The practical lever: move UPF from a typical 50–60% of the diet to under 30%, starting with sweetened drinks, snacks, and sauces.

What you'll learn

  • The operational test separating NOVA Group 3 from Group 4 — the line that matters.
  • What Hall's 2019 NIH trial proved, and why it's the strongest causal evidence in nutrition.
  • The dose-dependent epidemiology — ~250 studies, harms independent of weight.
  • The four engineering levers UPF uses to bypass appetite regulation.
  • Why reformulation is structurally incapable of fixing the category.
  • A practical target — under 30% of calories from UPF, and where to cut first.

NOVA in depth

Carlos Monteiro published the NOVA classification in 2009 in Public Health Nutrition (DOI 10.1017/S1368980009005291). It places food on a processing axis rather than a nutrient axis — explaining why a homemade pizza and a frozen supermarket pizza can be nutritionally identical and metabolically opposite. Four groups:

  • Group 1 — Unprocessed or minimally processed. Fresh, dried, frozen, fermented, or pasteurized whole foods: apples, brown rice, lentils, eggs, plain yogurt, frozen broccoli, ground beef, rolled oats.
  • Group 2 — Processed culinary ingredients. Substances extracted from Group 1 or nature, used to cook and season: oils, butter, salt, sugar, vinegar, honey.
  • Group 3 — Processed foods. Group 1 foods preserved or modified with Group 2 ingredients using methods a home cook recognizes (canning, fermentation, smoking, curing, baking): real bread, tinned tomatoes, smoked fish, aged cheese, cured ham, sauerkraut, olives in brine.
  • Group 4 — Ultra-processed food. Industrial formulations of substances extracted from foods (RBD seed oils, modified starches, hydrolyzed proteins, glucose syrups, maltodextrin, invert sugars) or lab-synthesized (artificial flavors, low-calorie sweeteners), held together with additives no home cook owns: emulsifiers, gums, stabilizers, flavorings, colors.

The line between Group 3 and Group 4 is the one that matters. A real sourdough and a "soft whole grain" supermarket loaf have similar macros, but the supermarket loaf contains DATEM, mono- and diglycerides, modified wheat starch, dough conditioners, and added gluten — ingredients a home baker has no source for.

The operational test, in van Tulleken's phrasing: if it contains an ingredient you wouldn't have in your kitchen — modified starches, RBD oils, protein isolates, emulsifiers, gums, flavorings, or low-calorie sweeteners — it's UPF. One such ingredient flips a product into NOVA 4. Organic Heinz baked beans are NOVA 3; the standard variety is NOVA 4 because of modified cornflour and spice extracts.

Lustig's parallel test in Metabolical arrives at the same line from biochemistry: absence of fiber and presence of added sugar, plus seed oils, emulsifiers, nitrates, and heat-generated toxins. Real food protects the liver and feeds the gut; UPF does neither. Marion Nestle calls UPF "the right unit of analysis" — what research and regulation should be organized around, not any single nutrient.

NOVA is arbitrary at the edges. That's fine — so is "alcoholic" vs "social drinker." A model is useful if it has predictive power, and NOVA does: it predicts mortality, CVD, cancer, T2DM, dementia, and IBD across dozens of cohorts after adjustment for nutrients.

The Hall 2019 trial

The strongest causal evidence in nutrition is Kevin Hall's 2019 inpatient trial at the NIH Clinical Center, in Cell Metabolism (DOI 10.1016/j.cmet.2019.05.008). Hall is a physicist; he was openly skeptical of NOVA and designed the trial to disprove it.

Twenty adults, NIH metabolic ward, four weeks. Two two-week phases in random crossover order: 80% UPF by calories, and 80% NOVA Group 1–3. Diets matched within tight tolerances on calories presented, sugar, salt, fat, fiber, and macronutrients. Subjects ate ad libitum — three meals plus snacks per day, roughly twice the calories they could reasonably eat.

On the UPF arm, subjects ate roughly 500 extra calories per day and gained about a pound a week. On the matched whole-food arm, the same subjects lost weight. The gap was visible by day three. Hunger and fullness ratings were similar across arms — subjects weren't reporting they wanted more food. They ate faster and ate more, unconsciously.

This rules out the common skeptical counters. The same people ate both diets, so income and lifestyle don't confound. Diets matched on sugar, salt, fat, and fiber should have produced equal intake if those were the relevant variables. They didn't. Processing itself drove the gap. Hall now treats UPF as the central variable in his lab.

The four engineering levers

If Hall's trial shows that UPF drives overconsumption, the question is how. Van Tulleken's framework identifies four engineering levers. None is sufficient alone; all four together are the modern UPF product.

Lever 1 — Destroyed food matrix. Fardet's food-matrix concept: physical structure, not just nutrients, determines how food interacts with the gut and brain. A 1977 study showed whole apples produced slow blood-sugar rise and lasting fullness; apple juice and puree produced spikes and crashes. UPF pulverizes the matrix and reconstructs it shelf-stable — soft (no chewing), dry (low water), energy-dense. On Hall's UPF arm, subjects ate roughly 17 more calories per minute than on the matched whole-food arm.

Lever 2 — Hyperpalatable salt-sugar-fat synergy. Howard Moskowitz's 1970s industry research identified the "bliss point" — the ratio at which a food is maximally palatable. Real foods rarely combine sugar, salt, and fat at high levels (steak is salt + fat; apple is sugar + fiber). UPF stacks them. Pringles deliver salt, fat, modified starch, and umami enhancers (glutamate, guanylate, inosinate) — sensory bait signaling aged protein the food doesn't deliver. Cola "speedballs" sourness (phosphoric acid), bitterness (caffeine), cold, and fizz to smuggle nine teaspoons of sugar past the body's sickly-sweetness limit. Warm flat cola is undrinkable; that's appetite control functioning.

Lever 3 — Mismatched taste signals. Dana Small at Yale showed humans learn to want flavors paired with calories — the body builds nutritional expectations from sensory cues. Low-calorie sweeteners break the contract: sweetness without predicted energy. Allison Sylvetsky's work showed children on diet drinks consume more total calories than children on water. The 2014 Suez et al. Nature study showed sucralose, aspartame, and saccharin altered the mouse gut microbiome and raised blood glucose — and the effect transferred via fecal transplant. Sweeteners are not metabolically neutral.

Lever 4 — Microbiome damage. Chassaing and Gewirtz's 2015 Nature paper (DOI 10.1038/nature14232) showed two common emulsifiers — carboxymethylcellulose (E466) and polysorbate 80 (E433) — at human-relevant doses thinned the gut's protective mucus, allowed bacteria to encroach on the epithelium, produced inflammation, and induced metabolic syndrome and (in susceptible animals) colitis. Tim Spector's son Tom lost roughly 40% of his detectable gut microbial species in ten days of fast food, with the loss persisting for years. Xanthan gum has colonized the microbiomes of billions of UPF eaters with a species absent in hunter-gatherers (Ostrowski, Michigan); trehalose has been linked to virulent C. difficile outbreaks. The microbiome produces 50,000 chemicals and regulates immune function; UPF reshapes it within weeks.

The epidemiology

Roughly 250 epidemiological studies now examine UPF intake against health outcomes (2,000+ in adjacent literature). The pattern is dose-dependent: every 10% increase in UPF share of intake produces a proportional rise in disease risk. From Sam Dicken's review (Batterham group, UCL), indicative numbers:

  • All-cause mortality: +10% UPF → ~+15% risk
  • Breast cancer: +10% UPF → ~+10% risk
  • Type 2 diabetes: +10% UPF → ~+15% risk
  • Dementia: +10% UPF → ~+25% risk
  • Cardiovascular disease, depression, IBD, frailty: all dose-dependent in the same direction

These effects persist after controlling for nutrient content, dietary pattern, smoking, alcohol, activity, income, and education — they are not artifacts of UPF eaters being poorer or unhealthier. They are also not mediated entirely by obesity: heart disease, dementia, IBD, and depression rise with UPF intake even in weight-stable people. UPF harms the body through inflammation, microbiome disruption, and metabolic dysregulation independent of weight gain. Even if you don't get fat, UPF is doing things to you.

Why reformulation doesn't fix it

The food industry's stable response to UPF criticism for four decades has been reformulation: lower the sugar, lower the sodium, add fiber, add protein, add vitamins. Lower-sugar Cheerios. "Healthier" Coca-Cola with stevia. Plant-based meat with pea protein isolate.

This is structurally incapable of fixing the category, because NOVA Group 4 is defined by ingredients, not macros.

A cereal of whole oats, water, and salt is NOVA 1. The same cereal with sugar lowered 30% but containing maltodextrin, soy protein isolate, natural flavorings, modified corn starch, tocopherols, BHT, and "color from concentrated vegetable juice" is still NOVA 4. The emulsifiers and flavorings carry the microbiome and matrix damage regardless of macros. The Hall result holds for "healthier" UPF too — matrix destroyed, calorie density high, flavors mismatched, microbiome-active additives present.

Worse, reformulation usually adds processing. Sugar reduction becomes artificial-sweetener addition. Fat reduction becomes gums and emulsifiers to mimic the lost mouthfeel. "Plant-based" meat is soy protein isolate, pea protein, methylcellulose, leghemoglobin, coconut oil, and flavorings — extreme NOVA 4 with maybe a better climate footprint than beef but no closer to food.

Nestle: reformulation legitimizes UPF rather than displacing it. Lower-sugar Cheerios is not progress toward food; it is the same product with better marketing. The corollary at the shelf: the absence of bad nutrients tells you very little. The presence of industrial ingredients tells you everything.

Practical action

The U.S. and U.K. get 50–60% of calories from UPF. Portugal sits at 10%. A reasonable target is under 30% — where the dose-dependent epidemiology flattens (the curve is steepest above 50% and gentler below 30%). Total abstinence is unnecessary and usually counterproductive. Reducing UPF share is the lever.

Where to cut first, in descending order of impact:

  • Sweetened drinks. Soda, energy drinks, vitamin water, sports drinks, sweetened plant milks and coffee drinks. The highest-yield single change. Replace with water, plain coffee, plain tea, plain sparkling water.
  • Packaged snacks. Chips, cookies, crackers, granola bars, "protein bars," candy — the lever-1 + lever-2 combination at its purest.
  • Sauces and dressings. Industrial ketchup, dressings, marinades load HFCS, modified starches, emulsifiers, flavorings. Make your own with oil, vinegar, salt, mustard, herbs.
  • Breakfast cereals. Replace with rolled oats, plain yogurt with fruit, or eggs.
  • Ultra-processed bread. Replace with real bread (sourdough, NOVA 3 bakery loaves) or skip.

Keep these — NOVA 3 or below: frozen vegetables and fruit (often more nutritious than out-of-season "fresh"); tinned tomatoes, beans, sardines; real bread, plain whole-milk yogurt, real cheese, plain pasta, dried legumes; pasture-raised eggs, meat and fish, olive oil, vinegar, salt, honey.

Lustig's six-word adaptation: protect the liver, feed the gut. Real food does both. UPF does neither.

FAQ

Is sourdough UPF? A traditional sourdough loaf — flour, water, salt, starter, time — is NOVA 3. A supermarket "sourdough" with industrial yeast, dough conditioners (DATEM, mono- and diglycerides), modified starches, and sourdough flavoring is NOVA 4. The ingredient list decides, not the label.

Are plant milks UPF? Almost all commercial plant milks are NOVA 4 — emulsifiers (gellan gum, sunflower lecithin), stabilizers, added sugars, refined oils, fortifying micronutrients. Plain soy milk with two ingredients exists but is rare. They may be reasonable for lactose-intolerant or vegan eaters, but they are not minimally processed and not nutritionally equivalent to dairy.

Are protein powders UPF? Yes, by definition. Whey, soy, and pea protein isolates are extracted fractions — exactly what NOVA 4 is built from. Most also contain flavorings, sweeteners, gums, lecithin. A scoop occasionally is fine; three shakes a day is a high-UPF diet wearing a fitness label.

What about "minimally processed" claims? Unregulated marketing in the U.S. and U.K. — no enforceable definition. Pret a Manger built a brand on "natural" while selling bread containing DATEM, L-cysteine hydrochloride, glycerol, and ascorbic acid. Use the ingredient list, not the front-of-pack.

Is bagged salad UPF? Plain bagged leaves are NOVA 1 — chlorinated wash and gas-flush are preservation, not formulation. A salad kit with dressing and crispy toppings in sachets is NOVA 4 because of the dressing and toppings. Eat the leaves with your own dressing.

Beyond Burger and Impossible Burger? Extreme NOVA 4 — soy protein concentrate, coconut oil, methylcellulose, pea protein isolate, soy leghemoglobin, natural flavorings. Insulin response similar to a beef burger; minimal fiber. They may reduce a meat-eater's climate footprint, but they are not health foods. Beans, lentils, tofu, and tempeh are NOVA 1–3 and metabolically very different.

Why isn't NOVA in U.S. or U.K. dietary guidelines? Industry lobbying. The 2025 U.S. DGAC declined to recommend reducing UPF on the grounds that RCTs were "too short" — a standard not applied to other recommendations. Brazil, France, Israel, Belgium, and Uruguay incorporate NOVA. The U.S. and U.K. don't, because doing so would name specific products and companies.

Sources

  • Monteiro, C. A. (2009). Nutrition and health. The issue is not food, nor nutrients, so much as processing. Public Health Nutrition, 12(5), 729–731. DOI: 10.1017/S1368980009005291.
  • Hall, K. D., Ayuketah, A., Brychta, R., et al. (2019). Ultra-processed diets cause excess calorie intake and weight gain: an inpatient randomized controlled trial of ad libitum food intake. Cell Metabolism, 30(1), 67–77. DOI: 10.1016/j.cmet.2019.05.008.
  • Chassaing, B., Koren, O., Goodrich, J. K., et al. (2015). Dietary emulsifiers impact the mouse gut microbiota promoting colitis and metabolic syndrome. Nature, 519, 92–96. DOI: 10.1038/nature14232.
  • van Tulleken, C. (2023). Ultra-Processed People: The Science Behind the Food That Isn't Food. Knopf. Summary of ~250 dose-dependent epidemiology papers and the four-lever mechanism framework.
  • Lustig, R. H. (2021). Metabolical: The Lure and the Lies of Processed Food, Nutrition, and Modern Medicine. Harper Wave. Operational definition of UPF and the protect-the-liver/feed-the-gut frame.
  • Nestle, M. (2025). What to Eat Now. Basic Books. UPF as the right unit of analysis; reformulation critique.
  • Spector, T. (2020). Spoon-Fed: Why Almost Everything We've Been Told About Food Is Wrong. Jonathan Cape. Tom Spector's ten-day fast-food experiment and the 40% microbial-diversity loss.

Related modules

Related glossary