GRN 1312 answers a safety question for one protein. It does not answer the question a cheese-analog formulator asks first: will this protein build a structure? The notice, dated 19 December 2025 and filed by NewMilkBuzz BV and Formo Foods GmbH, covers recombinant bovine alphaS1-casein made in E. coli (DSM 35603) for use “up to 25g per 100g in cheese analogs”, according to FDA’s inventory entry. That entry lists FDA’s letter as “Pending”.

The filing also reports that no phosphorylation was detected in the recombinant protein, while the native control was phosphorylated. Peer-reviewed work on dephosphorylated milk casein suggests that casein function depends on phosphorylation. That work did not test this protein, so what follows is an inference and not a finding of failure. The question is open, and the notice does not close it.

The regulatory scope of GRN 1312, including what “Pending” means for a launch, is covered in our piece on withdrawn and pending recombinant casein notices and in the route-to-market analysis. This article does not repeat that work. It reads the dossier as a spec sheet: composition, phosphorylation, intake maths and the functional evidence offered. We could read only part of the notice, as the last section explains.

What the dossier defines: a protein powder, not a dairy system

The notified substance is a powder specified at a minimum of 90 g protein per 100 g (Dumas method, N x 6.25), of which more than 97% is the recombinant alphaS1-casein, measured by HPLC-UV. The sequence is the B variant of mature bovine alphaS1-casein (199 residues) plus an N-terminal methionine left by bacterial expression, giving 200 residues. The notice gives three non-consecutive batches.

Parameter Specification (Table 3) FFM670 FFM681 FFM689
Crude protein, g/100 g (N x 6.25) at least 90 96.54 93.16 95.31
alphaS1-casein, % of protein above 97 above 97 above 97 above 97
Moisture, g/100 g 5 or less 1 0.8 0.3
Ash, g/100 g under 5 0.12 2.43 1.19
Total fat, g/100 g 2 or less 0.22 0.15 0.16
pH none set 3.68 4.18 4.17
Calcium, mg/kg none set 8.8 8.0 7.2
Phosphorus, mg/kg none set 150 3,300 3,500
Sodium, mg/kg none set 260 9,200 4,200
Endotoxin (LAL), EU/mg none set 6 96 105
Phosphorylation none set none detected none detected none detected

Source: notice Tables 3, 4, 6 and 7, and Part 2.1.2.1. The spec table in Part 2.4 has no line for calcium, phosphorus, endotoxin or phosphorylation. Those values appear only as batch data or narrative.

Three things stand out for a formulator. First, the powder is not a caseinate. Calcium is under 10 mg/kg in every batch, and the batch pH values of 3.68 to 4.18 are consistent with the notice’s description of a protein precipitated at pH 4 with phosphoric acid. Any calcium the recipe needs has to come from somewhere else.

Second, the batches are not tightly matched on salts. Sodium ranges from 260 to 9,200 mg/kg and ash from 0.12 to 2.43 g/100 g. Ash sits inside its limit, and sodium has no limit at all.

Third, the 97% purity claim is an HPLC-UV figure, and the notice uses LC-MS/MS for identity only. For how crude-protein figures behave with microbial protein, see our explainer on crude protein versus true protein.

One arithmetic point, ours and not the notice’s: the reported endotoxin range of 6 to 105 EU/mg is 6,000 to 105,000 EU/g of ingredient, so a cheese analog at the 25% maximum would carry about 1,500 to 26,250 EU/g.

Practical consequence: treat the spec as a safety and identity spec. Ask the supplier for the parameters that drive behaviour in a cheese matrix, namely calcium, total and bound phosphate, sodium and the batch pH, and set your own incoming limits.

The phosphorylation line: one sentence, with limits

The notice’s identity section (2.1.2.1) states: “No phosphorylation was detected in the recombinant αS1-casein protein across any of the analyzed batches of αS1-CN.” Native bovine alphaS1-casein, by contrast, is described elsewhere in the notice (2.1.2.2) as phosphorylated at eight sites, and the notice’s control showed phosphorylation “at three of the reported phosphorylation sites”. The notice adds: “No additional phosphorylation sites were assessed.”

Read the limits carefully:

  • The method was LC-MS/MS peptide fingerprinting after trypsin and AspN digestion. Sequence coverage ranged from 93.5% to 99.0% across the three batches and was 99.5% for the control. In the pages we could read, the notice does not say whether each phosphorylation site fell inside the covered stretch.
  • The control was a vendor-supplied native alphaS1-casein (Sigma, CAS 9000-71-9). Three of its eight sites were observed, so the assay did detect phosphate where it was present, at least at those positions.
  • Total phosphorus in the ingredient ranges from 150 to 3,500 mg/kg, a 23-fold spread across batches. If the LC-MS/MS result holds in all three, the extra phosphorus in the two high batches is unlikely to be protein-bound phosphoserine. Total phosphorus is therefore no proxy for phosphorylation in this product, in either direction.
  • The notice’s SDS-PAGE comparison reports the recombinant protein as equivalent to the native one in that assay, and notes that caseins migrate more slowly than expected because of their large hydrodynamic size. It does not present the gel as a phosphorylation test, and the calculated mass it quotes is 22.98 kDa unphosphorylated versus 23.6 kDa with eight phosphates.

The notice also supplies the reason this matters, without connecting it to its own finding. Part 6.1 states that “Caseins belong to a large family of serine-rich phosphoproteins” and places alphaS1-, alphaS2- and beta-caseins mainly inside the micelle, “whereas their phosphoserine domains interact with calcium phosphate nanoclusters”. Elsewhere it calls the recombinant protein “substantially equivalent (i.e., comparable in structural and physicochemical properties)”. The equivalence tests we saw were LC-MS/MS peptide fingerprinting, SDS-PAGE, a thermostability assay and an in vitro digestibility assay. None of those measures calcium binding, micelle assembly or gelation.

Practical consequence: “substantially equivalent” in a GRAS notice is an argument about safety by sequence. It is not a claim that the protein binds calcium phosphate. Do not carry it into a formulation brief as one.

The intake maths: a worst-case ceiling

The exposure assessment uses NHANES 2021 to 2023 consumption data and assumes cheese analogs at the maximum 25 g per 100 g. It applies that level to conventional cheese and imitation cheese food codes, on the assumption that analogs are eaten to the same extent as their conventional counterparts.

Population Share of consumers Per-user mean, g/day Per-user 90th percentile, g/day Per-user mean, g/kg bw/day Per-user 90th percentile, g/kg bw/day
US, 2+ years 38.7% 7.37 17.00 0.12 0.27
Children, 2-5 42.1% 5.56 10.63 0.32 0.53
Children, 6-12 33.0% 6.39 15.89 0.19 0.45
Adolescents, 13-18 29.2% 7.30 17.66 0.11 0.26
Adults, 19+ 40.2% 7.59 17.67 0.09 0.22

Source: notice Tables 9 and 10. Consumer shares are from the g/day table. The g/kg table gives slightly different shares for two groups (42.4% for children 2-5 and 29.3% for adolescents).

Working back from the notice’s own numbers, a per-user mean of 7.37 g/day at a 25% inclusion implies about 29 g of cheese-type food a day, and the 17.00 g/day figure implies about 68 g. That is our arithmetic, not the notice’s. The notice itself says the estimate “is an overestimation” because it assumes all cheese is analog at the ceiling. It also argues that casein intake will not rise, because the ingredient substitutes for animal casein.

Two wrinkles matter to a formulator. The use level is worded two ways. The notice’s Part 3.1 says “a maximum level up to 25 g of αS1-casein per 100 g in cheese analogs”, while FDA’s inventory entry says the substance is intended “as an ingredient at up to 25g per 100g in cheese analogs”. The ingredient is specified at 90% protein or more, so on spec minimums a recipe at 25 g of powder per 100 g carries at least about 21.8 g of alphaS1-casein by our calculation, because 0.90 times 0.97 is 0.873. The difference between “ingredient” and “protein” is a few grams per 100 g, which is material at this level.

The notice also says αS1-CN “is intended as a replacement for animal-derived casein in cheese analogs”. An analog that replaced all casein with this single casein would be a very different protein system from milk. The notice’s own figures put alphaS1 at about 40% of bovine casein, with alphaS2 about 10%, beta about 35% and kappa about 12%.

Practical consequence: the intake figure sets the safety ceiling. It says nothing about what share of a cheese’s protein this ingredient can carry and still hold together. Use-level caps bite in practice, as the OneGo Bioalbumen waffle case shows, but a safety cap is not a performance window.

Technical effect: one sentence, no data

Part 2.7, the section FDA’s rules reserve for physical or technical effect, is a single sentence: the ingredient “functions as a structural and nutritional protein ingredient”, “contributing to the texture, body, and protein content of cheese analogs.” In the pages we could read, no rheology, no curd test, no melt or stretch measurement and no comparison to dairy casein accompany it.

The comparator is GRN 1145, Imagindairy’s beta-lactoglobulin notice of 11 May 2023. Its Part 2 carries the same technical-effect heading, but its subsections are identity, manufacturing, and product specifications and batch analyses, with none on technical effect. It describes the ingredient as a “non-animal protein source” for foods that currently use milk or plant protein, at a maximum use level of 35% in nutritional bars. We have read the notice, not FDA’s response to it. The pattern is that GRAS notices fix identity, specification and exposure, and function is left to the supplier.

Formo’s commercial claims go further than its filing. FoodNavigator reported on 2 June 2026 that Formo said “its recombinant casein is responsible for food product texture, structure, melt, stretch, emulsification and gelation.” vegconomist, the same day, wrote that “Casein is the principal structural protein in milk, responsible for texture, melt, stretch, emulsification, and gelation in dairy-based foods.” The second sentence describes casein as a class. The first attributes the behaviours to Formo’s protein. Neither article cites data, and neither is evidence of function.

Practical consequence: a sampling request should ask for function data, because the regulatory file contains none. For product-development guidance on this class of ingredient, see our formulation guide to fermentation-derived protein. For a parallel case where the notified use levels and a gel application are separate matters, see the Verley GRN 1241 analysis.

What the published work on dephosphorylated casein suggests

One peer-reviewed paper we could verify bears on the open question. Che and colleagues (Food Hydrocolloids, 2025) reassembled micelles from four milk caseins (alphaS1, alphaS2, beta and kappa) at three phosphorylation degrees, produced by enzymatic dephosphorylation. Micelle formation and calcium binding rose with phosphorylation, “fully dephosphorylated CNs hardly formed CN micelle structures”, and the fully dephosphorylated set “completely failed in gelation” under acidification, instead precipitating at its isoelectric point, pH 5.5. The authors framed the work around “microbial CNs that are expected to exhibit non-native phosphorylation levels”. The paper used milk caseins, not a bacterially made one.

Formulator’s question What GRN 1312 reports What the verified published work suggests
Is the protein phosphorylated? None detected in three batches; three of eight native sites seen in control Phosphorylation degree governs micelle assembly and calcium binding (Che 2025)
Will it form or join micelles? Not tested in the pages we read Fully dephosphorylated milk caseins hardly formed micelles (Che 2025)
Will it gel with acid? Not tested in the pages we read Fully dephosphorylated milk caseins did not gel and precipitated at pH 5.5 (Che 2025)
Melt and stretch? Claimed in press, not in the notice pages we read The Che abstract does not mention melt or stretch

The inference is narrow. If function in cheese depends on casein phosphorylation, as Che suggests for milk-derived casein, a protein with none is a risk for micelle-dependent structure. It is not proof. That paper does not test this protein, this process, a single-casein system at 25 g per 100 g, or a recipe built around something other than a micelle. The dephosphorylation inference is ours, drawn from milk casein, and the evidence does not show that GRN 1312’s product fails.

The counter-argument: notices are safety documents, and Formo may not be selling one casein alone

The strongest reply is that a GRAS dossier is not a technical data package. A company may hold function data and keep it out of a public safety filing. It owes FDA no rheology, and the absence of such data from the notice is not evidence the data do not exist.

A second reply is formulation. A product need not rely on alphaS1-casein alone. It could be combined with other caseins, with dairy casein in a hybrid, or with a plant or hydrocolloid network, and a cheese analog need not be a micelle-based gel. FoodNavigator quotes Formo’s CEO on “continued development of more functional and customized casein variants”. That reads as a roadmap for other variants, which may or may not differ in phosphorylation, but it is a statement of intent and no data accompany it.

Third, the paper cuts both ways. Che reported that beta and kappa casein alone at native full phosphorylation were unstable at the salt concentrations used, so phosphorylation is not a sufficient condition for a stable system either.

The counter-argument fails only on one point: it does not show that this ingredient, at the use levels the notice allows, builds a cheese structure. That has to come from data.

What we could not establish

  • Function of the recombinant protein. No micelle, curd, melt or stretch data for GRN 1312 appear in the notice pages we read or in the press coverage.
  • The whole notice. Our fetch tool returned only part of the PDF; it was cut off in Section 6.1, so Sections 6.2 onward, the references and the appendices were not read. The page count differs between the brief (94) and the tool (70), and we cannot reconcile them.
  • Notice claims we could not find in the readable text. An LC-MS/MS abundance figure for alphaS1-casein and a host-protein count, and any EU/g endotoxin figure for cheese analogs. They may sit in the unread sections or not exist; the earlier draft’s figures for them are removed.
  • Which phosphorylation sites were covered. The notice reports coverage of 93.5% to 99.0% and, in the text we read, does not say which residues were missed.
  • Why phosphorus differs 23-fold between batches. The notice does not attribute it in the pages we read.
  • Antuma et al. (2023, Food Research International). The draft cited this paper, but we could not retrieve its abstract or confirm its citation details, so no claim from it is made.
  • Whether Formo pairs this ingredient with other caseins. The notice covers alphaS1 alone, and the press describes variants without detail.
  • FDA’s view. The inventory entry shows a letter status of “Pending”. We have no agency text addressing structure.

What a formulator should ask for

Ask for a supplier statement on phosphorylation by site and method. Ask for a cheese-matrix study at 25 g per 100 g that reports micelle or particle size, calcium partition, rennet or acid gel strength, and melt and stretch against a dairy-casein control. Ask what the recommended calcium and phosphate system is. Ask whether the product is sold with other caseins, and whether the next variants carry phosphate.

What to watch

Each of these can be tested against public information:

  1. A technical datasheet or application note with cheese data. Does it report gel strength, melt and stretch versus a dairy control, or only protein content?
  2. A phosphorylation specification. Does a revised spec or any amendment add a phosphorylation parameter, or does Formo’s “customized casein variants” language lead to a phosphorylated product?
  3. FDA’s reply on GRN 1312. The letter field will move off “Pending” when FDA responds. Check whether the letter or the file says anything about technical effect.
  4. Peer-reviewed data on bacterially made casein. A paper on recombinant, not enzymatically stripped, casein that tests the micelle and rennet questions would settle more than inference can.
  5. Customer launches. Formo was reported as sampling with US customers in June 2026. Launch claims for hard cheese would show whether the structure question was solved by formulation.