Does Processing Destroy Food Allergens? What Codex CXC 80-2020 Actually Says

No food process reliably eliminates allergenicity. Heat, acid, enzymatic hydrolysis, fermentation, high pressure and the newer non-thermal technologies all modify allergenic protein. Sometimes they reduce IgE binding substantially, sometimes they change nothing measurable, and in documented cases they increase allergenicity or create new sensitising agents.

Codex says so in terms:

“Allergens need to be managed throughout the supply chain and production process. Treatments lethal for pathogenic microorganisms, such as heating, high pressure processing, etc. generally do not destroy allergenic proteins. Processes that degrade proteins, such as enzymatic or acid hydrolysis, should not be relied upon to eliminate or completely destroy allergenic proteins.”

— Codex Alimentarius, Code of Practice on Food Allergen Management for Food Business Operators (CXC 80-2020, adopted 2020, amended 2026), Introduction

So “it goes through a cook step, so the allergen is destroyed” puts you against the express position of the code the GFSI-benchmarked schemes are built on. There is a second failure mode too: processing degrades the method you rely on to detect the allergen. A heat-processed, hydrolysed or Maillard-modified matrix can under-report severely on a standard ELISA — the certificate of analysis reads “not detected” and the protein is still there.

Learn to do this properly: Advanced Allergen Management & Validation — R3,450, approximately 40 hours, self-paced, with a full module on allergen removal, degradation and inactivation. All courses.

Key facts: what processing actually does to an allergen

Process	Effect on allergenicity	Effect on the test method	Removes a declaration obligation?
Cooking, baking, pasteurisation, retorting	Variable. Most major allergens are heat-stable; denaturation disrupts conformational epitopes, not linear ones.	Degrades ELISA recovery; under-reporting likely.	No
Dry roasting (peanut)	Increases IgE binding and stability via Maillard modification of Ara h 1, Ara h 2, Ara h 3.	Glycated protein extracts differently; recovery must be re-established.	No
Boiling	Can fall by leaching protein into the water — relocation, not destruction.	Cooking water is now an allergen-bearing stream.	No
Enzymatic or acid hydrolysis	Reduces IgE binding with degree of hydrolysis; linear epitopes remain. CXC 80-2020: “should not be relied upon”.	Defeats sandwich ELISA. Needs competitive ELISA or LC–MS/MS.	No
Fermentation and microbial proteolysis	Variable and product-specific; demonstrated, never assumed.	Fragmented protein degrades immunoassay performance.	No
HPP, PEF, ultrasound, cold plasma, UV, pulsed light, irradiation, extrusion, transglutaminase cross-linking	In-vitro IgE-binding reductions under particular conditions. Not standardised.	Affects extraction and recognition.	No
Refining, bleaching, deodorising (oils and other derivatives)	Strips protein out of the oil — removal, not modification.	Protein genuinely absent; DNA methods may still detect traces.	Only where a competent authority has exempted it under CXS 1-1985 4.2.1.6, on a risk assessment

What does Codex say about processing and allergens?

Codex states the position directly. CXC 80-2020’s Introduction says that treatments lethal for pathogenic microorganisms, such as heating, high pressure processing, etc. generally do not destroy allergenic proteins, and that processes which degrade proteins, such as enzymatic or acid hydrolysis, should not be relied upon to eliminate or completely destroy allergenic proteins.

Where it sits. The passage is in the Introduction to CXC 80-2020 — the final paragraph of the unnumbered opening portion, immediately before the sub-heading “Hazard characterization”. It carries no section number. Cite it as the Introduction; anyone who hands you a clause number has invented one.

What CXC 80-2020 is. A code of practice, not a certification standard. Its dominant modal is “should”, it is guidance, and no auditor raises a non-conformity against it directly. What it does is set the international reference position that the GFSI-benchmarked schemes build on. See our companion piece on BRCGS Issue 9 clause 5.3 and allergen audit readiness.

What CXC 80-2020 does not contain

CXC 80-2020 contains no reference doses, no thresholds and no action levels — those terms do not appear in it — and no precautionary allergen labelling provisions at all. “Precautionary” does not appear in the code, so sourcing “may contain” practice from it cites the wrong document: PAL and the reference doses live in the CXS 1-1985 annex, as Table A1 for IgE-mediated food allergy and Table A2 for coeliac disease (see our VITAL 4.0 precautionary labelling guide). The code has no allergen list either — it points to Sections 4.2.1.4 and 4.2.1.5 of CXS 1-1985 — and no HACCP section. Risk assessment appears only as a principle: controls “should be risk-based”.

How to quote the Codex processing passage correctly

Over-reading this passage damages your credibility as much as ignoring it. Codex says lethal treatments “generally” do not destroy allergenic proteins, not “never”. It says protein-degrading processes “should not be relied upon” — a recommendation against reliance, not a prohibition on hydrolysis. State your position at exactly that strength and no higher.

Three points of precision, each routinely mangled:

1. “Generally do not destroy” is not “never destroy”. Codex describes the general case and leaves room for exceptions. Say that Codex rules out heat destroying an allergen and you have overstated it.
2. “Should not be relied upon” is a recommendation against reliance. Not a claim that hydrolysis never reduces allergenicity — it plainly does, which is why extensively hydrolysed infant formula exists — and not a prohibition on hydrolysis. What it forecloses is hydrolysis as the control.
3. The “etc.” is in the source, so the list of lethal treatments is open and the reasoning reaches technologies the code does not name. And the object is “allergenic proteins” — not allergenicity in the abstract, and not detectability. Keep the noun.

The correct formulation is therefore not “processing does nothing”. It is: no thermal, hydrolytic, fermentative or novel process may be relied upon as an allergen control, and none substitutes for segregation, validated cleaning and accurate labelling. That is defensible against the text; anything stronger is not.

Does cooking destroy food allergens?

No. Cooking does not destroy food allergens as a class. Most major food allergens are heat-stable to some degree, and many of the proteins responsible for severe reactions survive normal cooking, baking, pasteurisation and retorting unchanged in clinical relevance. Heat modifies protein structure; it does not reliably remove allergenicity.

The reason sits in epitope biology. Conformational (structural) epitopes depend on the folded three-dimensional shape, so heat denaturation can disrupt them. Linear (sequential) epitopes are a stretch of amino acid sequence; denaturation does not touch them, and unfolding can even expose buried ones.

Allergens dominated by linear epitopes stay clinically active after cooking, and that is no fringe category. Thermostable families include the seed storage proteins (2S albumins, 7S/11S globulins — peanut, tree nuts, sesame), non-specific lipid transfer proteins, caseins, ovomucoid, parvalbumin and tropomyosin. Epitope type is not the whole mechanism: the 2S albumins, the lipid transfer proteins and ovomucoid also sit on compact disulphide-stabilised folds that resist irreversible unfolding, so their conformational epitopes survive heating as well. Never infer that an allergen is heat-labile because its epitopes are conformational.

Two patient-level observations are routinely misquoted in factories. PR-10 proteins and profilins of pollen–food syndrome are heat-labile, which is why some patients tolerate cooked apple, carrot or hazelnut; and extensive baking in a wheat matrix renders milk and egg tolerable for a subset of allergic children. Both are clinical findings about diagnosed patients under medical supervision — not a process control, a labelling basis or a reason to omit a declaration.

Removal, reduction, destruction of the analyte — the distinction that changes everything

Three completely different things happen when a process acts on allergenic protein, and blurring them is the root of almost every bad allergen decision in industry. Only one of the three can support a non-declaration decision, and only with validation behind it.

1. Removal — physically taking allergenic protein out: cleaning, refining, extraction, filtration, air capture. The only route that can support a non-declaration decision, and only where validated. Our guide to allergen cleaning validation covers the plant-hygiene side.
2. Reduction of allergenicity — modifying protein so less IgE binds. Real ingredient science: hypoallergenic formula, low-allergen cultivars, hydrolysates. Not a plant hygiene control.
3. Destruction of the analyte — degrading protein so the assay stops seeing it while clinically active fragments remain. A measurement failure mode masquerading as a control.

Number three is where sites get hurt: it produces exactly the evidence a complacent quality system wants, a clean result. The protein has not moved; the antibody has stopped recognising it.

Does roasting reduce peanut allergenicity?

No — dry roasting increases peanut allergenicity. Roasting drives Maillard and glycation chemistry that modifies the major peanut allergens Ara h 1, Ara h 2 and Ara h 3, increasing IgE binding and protein stability relative to raw or boiled peanut. Roasting is the canonical demonstration that heat is not an allergen reduction step.

Ara h 2 and Ara h 6 are the peanut markers most associated with severe reactions, and roasted peanut is the form most commonly consumed. Glycated protein also extracts differently and is recognised differently by antibodies raised against native protein.

The practical instruction is almost universally ignored: if your process history includes roasting, baking, extrusion or any browning step, re-establish analytical recovery in that matrix. CXC 80-2020 Section 6.5 says the same — a test “should be validated to work with the matrix/food of concern”. That discipline is taught step by step in the ASC advanced allergen validation programme.

Does boiling remove allergens?

Boiling can lower measurable allergen levels in some commodities, but largely because soluble protein leaches into the cooking water. The protein has been relocated, not destroyed — and the water is now an allergen-bearing stream that must be routed, controlled and kept out of product.

That is the governing principle for every removal step: if protein leaves the system, it has to go somewhere. Rinse water, filter cake, dust, spent oil, condensate, reclaimed detergent — each is now a hazard. Codex says as much, advising manufacturers to evaluate allergen cross-contact from cooking media such as water or oil.

Mid-article checkpoint. If any of this is new, your hazard analysis probably contains an unvalidated assumption about processing. The Advanced Allergen Management & Validation course exists to close those assumptions — R3,450, nine modules, a 90-question proctored exam at 80%.

Does hydrolysis make a protein safe?

No. Enzymatic hydrolysis reduces IgE binding roughly in proportion to the degree of hydrolysis, but residual peptides carrying linear epitopes remain. Extensively hydrolysed infant formula can still provoke reactions in a minority of cow’s milk allergic infants — which is precisely why amino-acid-based formula exists for the most sensitive patients.

Hydrolysis is genuine allergenicity-reduction science; partially hydrolysed (pHF) and extensively hydrolysed (eHF) formulas are made on that basis. But reduction is not elimination, and the direction of travel is not always downward.

The hydrolysed wheat protein case

The standing caution is the documented cluster of wheat-dependent allergy and anaphylaxis associated with hydrolysed wheat protein used in cosmetic products in Japan. The hydrolysate sensitised people who had previously tolerated native wheat. Modification created a new sensitising agent — which dismantles the intuitive model. Processing is not a slider that only moves toward “less allergenic”; it is a chemical transformation with novel outputs that need their own assessment.

Hydrolysates, fermentation and AN-PEP

• A sandwich ELISA needs two antibodies to bind two intact epitopes on one fragment. Hydrolysis destroys that geometry, so a sandwich ELISA “negative” on a hydrolysate is not a negative result but an uninterpretable one. Use a competitive ELISA or peptide-marker LC–MS/MS.
• Fermentation. Sourdough and lactic fermentation can degrade gluten substantially, and extended fermentation with proteolytic cultures reduces measurable allergen in some products, long-brewed soy sauce among them — variably, and demonstrated per product. Fermented wheat products are nonetheless not safe for coeliac disease unless they meet the gluten-free threshold by validated measurement.
• AN-PEP (Aspergillus niger prolyl endopeptidase) degrades gluten peptides in the stomach and is studied as a digestive aid for incidental exposure. It is not a means of making food gluten-free and must never be presented as one for coeliac disease. Codex defines coeliac disease by reference to gluten proteins from wheat, rye, barley and triticale.
• Strong acid or alkaline hydrolysis does cleave peptide bonds and destroy most epitopes — at ingredient-manufacturing conditions, not conditions a food process reaches incidentally.

Why a processed matrix can under-report allergen severely

Processing that denatures, hydrolyses or glycates protein degrades ELISA recovery, so a heavily processed sample can return a result far below the protein actually present. A “not detected” on a method never validated in that matrix is not evidence of absence — it is an absence of evidence.

Worked example — a recovery study that reverses a release decision

A retorted, high-fat ready meal is checked for milk protein carry-over on a shared line. The laboratory runs a spike-recovery study in that matrix first.

1. Spike level: 10.0 mg/kg total milk protein into the processed matrix.
2. Measured on the spiked sample: 3.2 mg/kg.
3. Recovery (%) = (measured ÷ spiked) × 100 = (3.2 ÷ 10.0) × 100 = 32%. Well outside the 50–150% window taught as acceptable guidance: 68% of the milk protein present is invisible to the assay.
4. Routine sample, raw result: 3.0 mg/kg (3.0 ppm).
5. Corrected = raw ÷ (recovery ÷ 100) = 3.0 ÷ 0.32 = 9.375 ppm.
6. Action level. The CXS 1-1985 annex expresses it as reference dose (mg) divided by amount of food consumed (kg), preferably at the 50th percentile. Codex Table A1 gives milk 2.0 mg total protein; reference amount 250 g. 2.0 ÷ 0.250 = 8.0 mg/kg. The VITAL 4.0 ED05 form — 2.0 ÷ 250 × 1000 — gives the identical 8.0 ppm.

The raw 3.0 ppm sits below the 8.0 ppm action level and the site releases. The corrected 9.375 ppm sits above it. The uncorrected number did not understate the level — it inverted the decision. (On the ED01 basis of 0.2 mg the action level would be 0.8 ppm, so even the raw result would have been actionable.)

Second worked example — the frameworks are not interchangeable

Correction cuts both ways, and so does the choice of reference dose. Take a 300 g portion assessed for gluten.

1. Codex Table A2 gives 4.0 mg of total gluten from all relevant sources for cereals containing gluten. Action level = 4.0 ÷ 0.300 = 13.33 mg/kg.
2. VITAL 4.0, from the Allergen Bureau, gives 5.0 mg as total protein of the gluten-containing cereal. Action level = 5.0 ÷ 300 × 1000 = 16.67 ppm.
3. A different number and a different quantity — never mix bases inside one assessment. On the VITAL 3.0 ED01 basis of 0.7 mg it would be 0.7 ÷ 300 × 1000 = 2.33 ppm.

Doing the arithmetic properly removes unnecessary “may contain” statements as often as it catches dangerous ones — but only if the framework is stated on the assessment.

What about HPP, pulsed electric field, cold plasma, UV and irradiation?

High pressure processing, pulsed electric field, ultrasound, cold plasma, UV and pulsed light, irradiation, extrusion, glycation conjugation and enzymatic cross-linking have all been reported to reduce IgE binding for particular allergens under particular conditions. These are allergenicity-reduction research findings relevant to ingredient design. They are not plant allergen controls.

• Results are allergen-specific and condition-specific; they do not generalise.
• Most are measured in vitro by IgE binding, not by clinical challenge. In-vitro reduction in IgE binding is not evidence of clinical safety.
• They are not standardised — there is no validated parameter set to operate to.
• HPP, one of the two treatments Codex names, has limited effect alone; it becomes more effective combined with enzymatic hydrolysis, because unfolding improves protease access. Codex’s list is open — “etc.” — so the reasoning reaches technologies it does not name.
• Low-allergen cultivars, RNA interference and gene editing hold promise but are not controls available today, and a silenced product would still require declaration.

What actually removes allergenic protein?

• Refined oils and other derivatives — read the exemption carefully. Refining, bleaching and deodorising do strip protein out of the oil, but the declaration consequence is not the operator’s to award. CXS 1-1985 4.2.1.6 provides that regional or national competent authorities may exempt ingredients derived from the listed allergenic foods from being declared, and that such exemptions “shall be subject to a risk assessment to establish the safety of the allergenic food derivative”. The standard never uses the phrase “highly refined” and names no exempted derivative. An exemption therefore exists only where a competent authority has granted one for your market. You cannot self-certify it from the commodity name, and cold-pressed, expelled and unrefined oils retain protein anyway.
• Washing and leaching, filtration and membrane separation, protein precipitation, starch purification, chromatographic removal — each with a reject stream to manage.
• Validated cleaning — the one an audit interrogates hardest. CXC 80-2020 Section 6.2.1 puts it precisely: the validation process “should be specific to the allergen, process and product matrix combination”.

Three method cautions. PCR is unaffected by protein denaturation because DNA persists — which cuts both ways: DNA can be present where protein is not, as in refined derivatives, and protein present where DNA is absent, so PCR never confirms protein. Lateral flow devices perform poorly on heat-denatured and hydrolysed protein and can hook-effect low at very high concentrations, so a negative strip on a processed matrix is weak evidence; confirm by quantitative ELISA or LC–MS/MS. And ATP bioluminescence and total-protein swabs measure general soil — cleaning indicators that can never validate allergen removal.

What this means for your allergen programme

1. Strike every “the process destroys it” justification unless it is a validated removal step with evidence. Quote CXC 80-2020’s Introduction in whatever replaces it.
2. Make process history an input to method selection. Hydrolysates need competitive ELISA or LC–MS/MS; processed and glycated matrices need recovery re-established by spike-recovery study in that matrix, with the correction applied to every result.
3. State the reporting basis and the reference-dose framework. Basis: total protein, commodity or specific marker — gluten as gliadin × 2 against the 20 mg/kg gluten-free threshold, peanut against Ara h 1, Ara h 2, Ara h 3 or Ara h 6, milk against casein or beta-lactoglobulin (Bos d 5). Codex’s caution: a casein test should not be used where whey is the allergen of concern. Framework: Codex Table A1 or A2, VITAL 4.0 ED05, VITAL 3.0 ED01 or a Netherlands ED05 value — never mixed.
4. Map every allergen-bearing stream a removal step creates, and control where it goes.
5. Keep the declaration. Under R146 of 2010, Regulation (EU) No 1169/2011 and the US “Big Nine” after the FASTER Act added sesame, an intentional allergenic ingredient is declared. No process step negotiates that away. Draft R3337 would tighten the South African rules but is not yet in force, and R638 of 2018 governs premises hygiene alongside it.

Where this is taught: Module 7

This is the seventh module of ASC’s advanced allergen validation course. It covers heat and epitope biology, hydrolysis and fermentation, the novel processes, and physical removal from air (capture at source, pressure cascade, recirculating air handling units), in CIP (TACT, the cool-to-warm pre-rinse before hot alkaline wash, spray shadows, reclaimed detergent carry-over, final rinse water as the assay point) and from machinery (design for disassembly, gaskets, valve internals, and high-pressure hosing as a cross-contact multiplier).

Frequently asked questions

Does cooking destroy food allergens?

No. Most major food allergens are heat-stable to some degree. Heat disrupts conformational epitopes but not linear ones, so allergens dominated by linear epitopes — seed storage proteins, caseins, ovomucoid, parvalbumin, tropomyosin — stay clinically active after cooking.

Does Codex say cooking destroys allergens?

The opposite. The Introduction to CXC 80-2020 states that treatments lethal for pathogenic microorganisms, such as heating, high pressure processing, etc. “generally do not destroy allergenic proteins”, and that protein-degrading processes such as enzymatic or acid hydrolysis “should not be relied upon to eliminate or completely destroy allergenic proteins”. The passage carries no section number; cite it as the Introduction.

Does roasting reduce peanut allergenicity?

No, it increases it. Dry roasting modifies Ara h 1, Ara h 2 and Ara h 3 through Maillard and glycation chemistry, raising IgE binding and protein stability relative to raw or boiled peanut.

Can a hydrolysed ingredient be declared as non-allergenic?

No. Hydrolysis reduces IgE binding but leaves residual peptides carrying linear epitopes, and in the hydrolysed wheat protein case it created a new sensitising agent. Codex advises that protein-degrading processes should not be relied upon to eliminate allergenic proteins. A sandwich ELISA on a hydrolysate is also uninterpretable — use competitive ELISA or peptide LC–MS/MS.

Why does a processed sample test lower than the allergen actually present?

Denaturation, hydrolysis and glycation change the protein the antibody was raised against, degrading ELISA recovery. A spike-recovery study in the actual processed matrix, with a correction factor applied, is the only way to know how much a result understates. CXC 80-2020 Section 6.5 says a test should be validated for the matrix of concern.

Is refined peanut oil exempt from allergen declaration under Codex?

Not automatically. CXS 1-1985 Section 4.2.1.6 allows competent authorities to exempt ingredients derived from listed allergenic foods, and such exemptions shall be subject to a risk assessment establishing the derivative’s safety. The standard never uses “highly refined” and names no exempted derivative, so an operator cannot self-certify one.

Build the competence before the auditor asks

Advanced Allergen Management & Validation: Scientific Principles, Risk Assessment and Cleaning Verification — R3,450, no VAT charged.

Nine modules, approximately 40 hours, online and self-paced, instructed by Mthokozisi Nkosi. Includes the ASC Allergen Validation Toolkit — an Excel workbook whose multi-framework risk assessment calculator switches between Codex Table A1/A2, VITAL 4.0 ED05, VITAL 3.0 ED01 and Netherlands ED05 reference doses, plus a swab recovery calculator, a result interpretation tool, an allergen matrix and a changeover risk assessment — and a document-controlled Cleaning Validation Protocol template. Assessment is a 90-question proctored exam at an 80% pass mark — 72 of 90 — in 150 minutes, with a certificate of competence on passing. FoodBev SETA accredited (587/00337/1900), HPCSA CPD accredited, SAATCA TC 065; aligned to FSSC 22000 Version 7, BRCGS Issue 9, IFS Food Version 8 and SQF Edition 9.

Enrol in Advanced Allergen Management & Validation →  |  Foundational allergen course  |  All courses

Related: Allergen cleaning validation guide  ·  VITAL 4.0 precautionary labelling  ·  BRCGS clause 5.3 audit readiness