Research note 008 — 2026-08-30 (expanded 2026-09-08)
Cheese Bitterness: Bacteria Write It, Then Erase It
That faint bitterness when you bite into aged cheese. It is not a single substance from milk, but a peptide fragment born partway through the enzymatic cutting-up of the protein casein. In other words, cheese's bitterness is 「生まれて、消えていく」 (in translation: "born, and then it disappears") — bacteria do both.
On this page (11 sections)
- What this page establishes
- The bitterness is a fragment of casein
- The difficulty of pinning the blame on a single peptide
- Receptors respond without regard to a peptide's origin
- Whether it turns bitter depends on the starter bacteria's "scissor habit"
- Changing the lineup of bacteria raises or lowers bitterness
- Debittering is a process that exists apart from ripening, too
- Bitterness can be predicted from hydrophobicity — a 50-year-old rule is contested
- Honest notes
- Frequently asked
- What is not known
What this page establishes
- That much of aged cheese's bitterness comes from bitter peptides formed by the breakdown of the protein casein has been confirmed across multiple studies.
- Trying to explain the strength of bitterness from a single peptide's concentration alone gives a weak correlation; an index combining multiple peptides has been reported to show a stronger correlation.
- Bitter receptors (such as TAS2R1) do not distinguish whether a peptide's origin is cheese; cell-based experiments have confirmed that they respond to any hydrophobic amino-acid sequence.
- That changing the type or combination of starter bacteria and yeast raises or lowers bitterness has been confirmed across multiple studies, but the mechanism is not simple — for instance, a bacterium's enzyme activity and transcription level do not necessarily move together.
The bitterness is a fragment of casein
In aged Gouda, 16 casein-derived bitter peptides were identified, and when synthesized and measured, their thresholds ranged 0.05-6.0 mmol/L. In Cheddar, simply adding the 5 main sequences back into a model cheese reproduced bitterness indistinguishable from the real thing. On the receptor side, identification has also progressed: bitter peptides strike TAS2R1 and others.
Scope: Aged Gouda and aged sharp Cheddar (water-soluble extracts).
Note: The explanatory power of any single peptide alone is limited: in Broadbent 2002, the correlation between a single peptide's concentration and the bitterness score stayed at r=0.56-0.58, versus r=0.84 for a combined 2-peptide index (see limits).
Show data
ingredient-cheese —derives_bitterness_in_part_from→ compound-class-bitter-peptidesScope: Peptides purified from aged Gouda cheese and then synthesized.
Note: The abstract gives neither the threshold-measurement method nor the panel size.
Show data
compound-class-bitter-peptides —has_recognition_thresholds_of→ "0.05-6.0 mmol/L (12 synthesized and purified peptides showed clear bitterness). The lowest threshold was 0.05 mmol/L, for the decapeptide YPFPGPIHNS, with a caffeine-like bitterness quality."Scope: Aged sharp Cheddar versus young mild Cheddar. A taste-reconstitution experiment.
Note: The sequences are as given in the abstract (residue numbers are not written because they are not given in the abstract).
Show data
compound-class-bitter-peptides —reconstitute_bitterness_of→ ingredient-cheeseScope: Heterologous-expression cell calcium assays: HEK293 (Maehashi 2008) and C6 glioma (Upadhyaya 2010)
Note: 2 independent lines of cell assay point to the same receptor. The contribution in vivo is not established.
Show data
compound-class-bitter-peptides —activates→ receptor-tas2r1The difficulty of pinning the blame on a single peptide
A 1996 review surveying cheese bitterness explains it as an inevitable consequence of proteolysis, and brings together, in a single volume, methods for isolating and detecting bitter peptides, their physicochemical properties, and even ways to control them (summarized within the scope of the abstract).
Scope: A review on cheese bitterness (1996). It covers isolation and detection methods for bitter peptides, their physicochemical properties, and control methods.
Show data
compound-class-bitter-peptides —is_produced_through→ "Cheese bitterness is described as an inevitable consequence of the proteolysis that proceeds during ripening, and the relationship between proteolytic activity and the accumulation of bitter peptides is organized in a review."Among the peptides that have been linked to aged Cheddar's bitterness is β-casein-derived f193-209. This peptide is generated by chymosin and remains uncut by the lactic-acid bacteria's proteinase (lactocepin). Even when the starter strain differed, its concentration stayed within a range of 0.3 to 0.9 mg/g, with no statistical difference observed. A separate study tracking aged Cheddar reports that this peptide's concentration showed a relationship with sensory-measured bitterness strength — a correlation coefficient of 0.803 at 180 days of ripening and 0.554 at 270 days (summarized within the scope of the abstract). The relationship loosens the more time passes, and the assessment is that it stands as a partial marker rather than the sole culprit.
Scope: Reduced-fat Cheddar cheese (Broadbent 2002). The association with bitterness is cited by that paper from earlier literature.
Note: The association with bitterness itself is cited by Broadbent 2002 from earlier literature (that paper's refs 4 and 28); there is no threshold measurement within this paper itself.
Show data
compound-class-bitter-peptides —includes_chymosin_derived_member→ "β-CN f193-209: a chymosin-derived peptide that has been associated with bitterness in cheese. It is not hydrolyzed by lactocepin, and its concentration shows no difference among starters (0.3-0.9 mg/g, P>0.5). Combined with αs1-CN f1-9, an index for predicting the tendency toward bitterness has been proposed (r=0.84)."Receptors respond without regard to a peptide's origin
Even though it is difficult to separate out the responsibility of each individual peptide, the receptor side offers a different view. Casein digested with trypsin for 12 hours and collected by gel filtration into a fraction (2.5 mg/mL) was recorded, by sensory evaluation, as being as bitter as 0.125 mM quinine. Viewed from the receptor side, four receptors responded to this fraction — hTAS2R1, hTAS2R4, hTAS2R14, and hTAS2R16 — with hTAS2R1 and hTAS2R4 showing the strongest responses. The fraction is not a single purified peptide but a mixture that includes nonbitter components as well — a result that shows the side of the fact that receptors get struck regardless of origin.
Scope: HEK293 heterologous expression, calcium imaging
Note: The fraction is a mixture of numerous peptides; the correspondence between individual peptides and receptors cannot be identified from this experiment alone (see limits).
Show data
compound-class-bitter-peptides —activates_multiple_members_of→ receptor-tas2r-familyThis same TAS2R1 is not the exclusive territory of cheese peptides. Oligoporins A, B, D, and E, compounds isolated from a species of mushroom, all activate TAS2R1 at a threshold concentration of 30 µM, while the closely related oligoporin F does not activate TAS2R1. This is an experiment in cells expressing the receptor, not a sensory evaluation of these compounds taken into the mouth. Whether the compound is a peptide or a mushroom-derived one, the cell-level fact that TAS2R1 responds and the sensory fact that a substance is felt as bitter sit at different layers.
Scope: Layer: cell. A screen using heterologous-expression cells for 26 human bitter taste receptors, with threshold concentrations measured by calcium imaging (Table 2).
Note: Oligoporins A, B, and E are represented as the same compound group by the oligoporin-D entity (no separate entities have been created).
Show data
compound-oligoporin-d —activates→ receptor-tas2r1Whether it turns bitter depends on the starter bacteria's 「ハサミの癖」 (in translation: "scissor habit")
Even with the same production method, the cleavage specificity of the lactic-acid bacteria's proteinase (lactocepin) makes a large difference in the bitterness score after 6 months. Some bacteria cut in a way that readily produces bitter peptides, while others break them down further and erase the bitterness — ripening is the simultaneous writing and proofreading of peptides by bacteria.
Scope: 50% reduced-fat Cheddar, a single isogenic starter, a trained panel (n=14).
Note: A result from a model system (reduced-fat, single-strain starter). Extrapolation to ordinary cheese is limited.
Show data
ingredient-cheese —develops_bitterness_depending_on→ "Specificity groups of Lactococcus lactis lactocepin (an extracellular proteinase). Against proteinase-negative-strain cheese (bitterness 1.8/9), groups a/e/h scored 3.5-4.7 (at 6 months, P<0.0001), with group h the most bitter."The very vigor with which bacteria use their scissors also changes with the medium's conditions. Growing Lactococcus lactis in a casein-based medium, PrtP peptidase activity rose the higher the protein and calcium concentrations, accompanied by an increase in casein-derived bitter peptide concentration. However, the transcription level of the prtP gene itself instead fell by more than 50% under conditions with 50 mM added calcium, and depending on the strain, activity fell 15% in one case and instead rose in another. Transcription level and actual enzyme activity do not necessarily move in the same direction.
Scope: Fermentation experiments using reconstituted casein media (protein 3.4% and 8.5% w/v) and L. lactis strains LTH 7122 and LTH 7123.
Note: At low calcium concentration (+5 mM), both prtP transcription and activity rose, but at high concentration (+50 mM) the responses of transcription and activity diverge — reported as a nonlinear relationship.
Show data
compound-class-bitter-peptides —is_produced_through→ "Casein breakdown by PrtP-positive Lactococcus lactis strains. Under medium conditions with higher protein and calcium concentrations, PrtP peptidase activity rose, accompanied by an increase in casein-derived bitter peptide concentration. However, the transcription level of the prtP gene itself instead fell by more than 50% under high-calcium conditions (+50 mM CaCl2), and the change in activity and the change in transcription level did not necessarily agree (depending on the strain, at +50 mM activity was observed to either fall by 15% or rise slightly)."This difference has been confirmed by actually making cheese. Fresh cheese from high-protein concentrated milk fermented with PrtP-positive lactic-acid bacteria strains recorded bitterness scores of 3.9 and 4.1 on a 0-to-5 scale, significantly higher than the 1.7 and 2.8 of cheese made with PrtP-negative strains (p<0.05). However, the PrtP-negative-strain cheese was still more bitter than a commercial fermented reference product (bitterness 0.5), and there is a mismatch, in the strength of the wording, between this study's conclusion summarizing that 「PrtP陰性株は苦味を生じさせなかった」 (in translation: "the PrtP-negative strains did not produce bitterness") and the numbers of the results themselves.
Scope: Experimental fresh cheese (CoF) made using reconstituted casein concentrated-milk media (protein 3.4% and 8.5% w/v). Sensory evaluation by a trained bitterness panel.
Note: This paper's conclusion section summarizes that "the PrtP-negative strains did not produce bitterness," but this is at odds, in the strength of the assessment, with the figures in the results section (see the corresponding evidence's limits line).
Show data
ingredient-cheese —develops_bitterness_depending_on→ "Fresh cheese made from high-protein concentrated milk fermented with extracellular-proteinase (PrtP)-positive Lactococcus lactis strains (LTH 7123, LTH 7122) showed bitterness scores of 3.9 and 4.1 (0 = no bitterness to 5 = extremely bitter; panel n≥12, i=2 replicates), significantly higher (p<0.05) than the 1.7 and 2.8 scores of cheese made with PrtP-negative strains (LTH 7117, LTH 7112). Cheese made with the PrtP-negative strains was also more bitter than a commercial fermented concentrated fresh-cheese reference product (FCo, bitterness 0.5). A control cheese acidified directly without fermentation (CoF) was rated at bitterness 1.0 (very slightly bitter), and there was no significant difference between CoF and FCo."Changing the lineup of bacteria raises or lowers bitterness
There is also the option of changing the combination of bacteria used in the starter itself. In reduced-fat Cheddar, using an integrated culture system that selects Lactococcus lactis subsp. cremoris SK11 as the primary acid-producing strain and combines it with two other strains is reported, within the scope of the abstract, to have given markedly lower bitterness intensity than using a commercial Cheddar starter.
Scope: Reduced-fat Cheddar cheese, ripened for 150 days. Chemical, microbiological, and sensory analyses were performed at regular intervals.
Note: The abstract does not give the specific score difference for "markedly lower," and the full text was not checked, so it is not written here.
Show data
ingredient-cheese —can_be_debittered_by→ "An integrated culture system combining 3 starter strains (Lactococcus lactis subsp. cremoris SK11, L. lactis subsp. lactis biovar diacetylactis JVI, and Lactobacillus casei 7A), with SK11 selected as the primary acid-producing strain. Cheese made with SK11 had markedly lower bitterness intensity than cheese made with a commercial Cheddar starter."Changing the type of bacteria used as an adjunct can also move things in the opposite direction. In an experiment conducted in South Africa, aged Cheddar to which either the yeast Debaryomyces hansenii or Yarrowia lipolytica was added alone is reported, within the scope of the abstract, to have shown earlier development of Cheddar-like flavor, but this was accompanied by the occurrence of bitterness as well.
Scope: An experiment conducted in South Africa. Aged Cheddar cheese compared across 4 groups: control, D. hansenii alone, Y. lipolytica alone, and both species combined.
Note: The specific sensory scale and panel size are not stated in the abstract.
Show data
ingredient-cheese —develops_bitterness_depending_on→ "Aged Cheddar cheese to which either Debaryomyces hansenii or Yarrowia lipolytica was added alone as an adjunct culture showed accelerated development of Cheddar-like flavor, but this was accompanied by the occurrence of bitterness as well."Cheddar made by incorporating a nisin Z-producing lactic-acid bacteria strain into the starter is also reported, within the scope of the abstract, to have shown both sourness and bitterness strengthened compared with the control, with both hydrophilic and hydrophobic peptides in the water extract increasing.
Scope: Cheddar cheese, aged 6 months. Sensory evaluation was carried out at the 6-month point.
Show data
ingredient-cheese —develops_bitterness_depending_on→ "Cheddar cheese made with a nisin Z-producing strain (Lactococcus lactis subsp. lactis biovar diacetylactis UL719) incorporated into the starter showed increased sourness and bitterness in sensory evaluation compared with control cheese. The same strain increased the generation of both hydrophilic and hydrophobic peptides in the cheese's water extract."On the other hand, combining a Lactobacillus casei adjunct with a nisin Z-producing strain is reported, within the scope of the abstract, to instead show a debittering effect, with the flavor evaluation also rising. Even with the same set of lactic-acid bacteria tools, depending on how they are combined they can turn out on either the side that adds bitterness or the side that removes it.
Scope: Same as above (Cheddar cheese, aged 6 months).
Note: The combination that produced 『脱苦味効果』 (in translation: "a debittering effect") (Lb. casei + nisin-producing strain) and the combination rated as having 『風味強度が最高』 (in translation: "the highest flavor intensity") (Lb. casei + liposome-encapsulated nisin) are distinguished in the abstract, and this note is written so as not to conflate them.
Show data
ingredient-cheese —can_be_debittered_by→ "Cheddar cheese to which an adjunct culture of Lactobacillus casei-casei L2A was added in combination with a nisin Z-producing strain is reported to have shown a debittering effect and improved flavor quality. Separately, cheese to which Lb. casei and liposome-encapsulated nisin Z were added was rated highest in flavor intensity in sensory evaluation and ranked first."Debittering is a process that exists apart from ripening, too
The work of removing bitterness also exists independently of cheese ripening itself. In a milk-protein hydrolysate used for cheese flavoring, using a protease that selectively cleaves the amino-acid residues Pro, Leu, Phe, and Val, the further hydrolysis was advanced (degree of hydrolysis raised from 7.5% to 28.0%), the more the bitterness sensory score dropped from 5.0 to 0.3-2.7, depending on the protease used. The bitterness of the whole mixture could not be explained by simply adding up the bitterness of individual peptides; the abundance of peptides with a molecular weight of 0.5 to 3 kDa and 5 to 23 residues had the strongest effect, reported within the scope of the abstract.
Scope: Cheese flavoring (an enzymatic hydrolysate) made from a milk protein (casein-based) hydrolysate. Sensory evaluation was combined with peptidomics (structural analysis).
Note: Note that the subject is not aged cheese itself but a "cheese flavoring" raw material produced by enzymatic hydrolysis.
Show data
compound-class-bitter-peptides —can_be_debittered_by→ "For milk protein hydrolysate (a cheese-flavoring raw material), extensive hydrolysis (degree of hydrolysis raised from 7.5% to 28.0%) by a protease that specifically cleaves Pro, Leu, Phe, and Val residues (with FU the most effective among several enzymes tested) lowered the bitterness sensory score from 5.0 to 0.3-2.7 (varying by the protease used). The bitterness of the whole mixture could not be predicted by a simple sum of individual peptides' bitterness; the analysis found that the abundance of peptides with a molecular weight of 0.5-3 kDa and 5-23 residues contributed most, followed by N-terminal basic amino acids and C-terminal hydrophobic amino acids."One of the tools responsible for this debittering process is microbial aminopeptidases. They are reported to lower bitterness by selectively cleaving, from the N-terminal side, hydrophobic amino acids exposed at the ends after protein has been cut up by endopeptidases; commercial aminopeptidases are reported, within the scope of the abstract, to be used not only for debittering protein hydrolysates but also for deep hydrolysis of protein and for making seasonings, cheese, and bioactive peptides.
Scope: A review on the properties and applications of microbial aminopeptidases.
Show data
compound-class-bitter-peptides —is_reduced_by→ "Microbial aminopeptidases. They are described as reducing bitterness by selectively cleaving, from the N-terminal side, hydrophobic amino acids exposed at peptide termini after endopeptidase hydrolysis of protein. The review states that commercial aminopeptidases are mainly used for debittering protein hydrolysates, deep hydrolysis of protein, and the production of seasonings, cheese, and bioactive peptides."「疎水性で苦味を予測できる」 (in translation: "bitterness can be predicted from hydrophobicity") — a 50-year-old rule is contested
The Q-rule (peptides with high mean hydrophobicity are bitter), proposed in 1971, is an empirical rule that has been used in food science for half a century, but measurements on soy peptides have explicitly refuted it, stating in so many words that 「Q値は苦味を支持しない」 (in translation: "the Q-value does not support bitterness"). NIGAMI LAB keeps both sides on record as contested.
Scope: Peptides. A prediction based solely on amino acid composition.
Note: Not supported in soy protein hydrolysates (Cho 2004, see contradicts). There is also a report of consistency at the dipeptide level (Maehashi 2008). The original Ney 1971 source was checked for bibliographic metadata only (metadata_only); the statement of the rule relies on a quotation in the full text of Maehashi 2008.
Show data
concept-q-rule —predicts_bitterness_from_mean_hydrophobicity→ "A peptide with mean hydrophobicity Q > 1400 cal/mol and a molecular weight in the range of about 100-6000 is bitter (Ney's 1971 empirical rule)."Tracing the original definition of this empirical rule reveals a somewhat finer line. In the earlier work the review cites, a line is drawn: peptides with a Q-value below 1300 cal/mol are not bitter, and peptides above 1400 cal/mol are bitter, with the range between 1300 and 1400 stated as outside the definition.
Scope: Layer: literature. A statement in a 2024 review that cites the original Ney 1971 source.
Show data
concept-q-rule —is_defined_as→ "In Ney's (1971) original statement of the Q-value rule (Ney's rule), a peptide with a Q-value below 1300 cal/mol is not bitter and a peptide above 1400 cal/mol is bitter; the range of 1300-1400 is not specified."A case that runs squarely against this rule can also be found in the earlier work the review cites. A study examining peptides obtained by autolyzing shrimp heads records that the bitter components concentrated in the 3-to-5 kDa molecular-weight fraction, and that the fraction with lower mean hydrophobicity conversely showed stronger bitterness. It also notes that hydrophobic amino-acid residues at the C-terminus contributed more strongly to bitterness than those at the N-terminus.
Scope: Layer: literature (secondary citation). A study of shrimp-head autolysate peptides by Fu et al. (original not read), as summarized by the Wang et al. 2026 review.
Note: Secondary citation. The original (Fu et al.) was not checked; this is based on the summary statement in Section 2.1 of Wang et al. 2026 (Foods, full text checked).
Show data
concept-q-rule —does_not_predict→ "A study examining peptides derived from the autolysis products of shrimp heads reports that the fraction with lower mean hydrophobicity showed, conversely, stronger bitterness; the bitter components were concentrated in the 3-5 kDa molecular weight fraction; and hydrophobic amino acid residues at the C-terminus contributed more strongly to bitterness than those at the N-terminus."Honest notes
- How far the primary literature has been checked differs by claim; some have been confirmed down to the full text, while others remain checked only at the abstract level (see each card's review-status display).
- All receptor findings are measurements from cells expressing the receptor in culture, not sensory evaluations of something taken into the mouth. The same holds for the oligoporins and for the casein-digest fraction.
- The counterexample to the Q-rule from shrimp-head digest peptides is based on the description in the earlier work the review cites, not on the original paper itself.
- 「苦味＝欠陥」 (in translation: "bitterness equals a defect") is a framing from the context of cheese quality control; a faint bitterness can instead be preferred as complexity in some cases (a question of preference is a separate matter).
Frequently asked
Why does cheese turn bitter?
Much of cheese's bitterness has been confirmed to come from bitter peptides formed when casein, milk's protein, breaks down during ripening. For example, more than a dozen bitter peptides have been identified from aged Gouda and Cheddar, and some of them produce a bitter taste even at very low concentrations. This breakdown is driven forward by enzymes carried by the starter bacteria, and it has also been confirmed that the type of bacteria and the specificity of the enzyme change how much peptide forms and how bitter it is. However, multiple studies state, as a limitation, that it is difficult to completely separate out, peptide by peptide, how much each one contributes to bitterness.
Is the 『Qルール』 (in translation: "Q-rule") that predicts bitterness correct?
The Q-rule, proposed in 1971, is an empirical rule that tries to predict whether a peptide is bitter from the hydrophobicity value (Q-value) of the amino acids that make it up, together with its molecular weight. Results consistent with this tendency have been reported for casein-derived bitter peptides and for some dipeptides, while a study using soy peptides has also reported a refutation, finding that bitterness does not simply correspond to hydrophobicity and instead shows a nonmonotonic relationship with molecular weight. In other words, the Q-rule is observed as a tendency in some systems, but it has not been confirmed as a rule that holds across peptides in general.
What is not known
- A method for completely separating out how much each individual bitter peptide contributes to the bitterness of the whole mixture has not yet been established.
- Why transcription level and peptidase enzyme activity diverge (transcription falls under high-calcium conditions while activity rises depending on the strain) has not been confirmed within the scope of this material.
- A mechanism explaining the boundary of why the Q-rule holds for some kinds of peptides and not others has not yet been measured.
- No material was found examining how much bitter substances that strongly activate a specific receptor, such as the mushroom-derived oligoporins, actually contribute to bitterness in real food.
Related: bitter peptides / TAS2R1 / Ney Q-rule / cheese / Research note 004: In Olive Oil, Bitterness Is Not a Defect / Research note 037: Fermentation Creates Bitterness, Then Breaks It Down — Miso, Soy Sauce, Sake / Research note 044: Who's to Blame for Sake's Bitterness? — The Peptide Account and the Amino-Acid Account
How to cite this page
Aoki, M. (NIGAMI LAB) (2026). Cheese Bitterness: Written, Then Erased. https://nigamilab.com/en/notes/cheese-bitter-lifecycle/
Each claim card has a stable anchor (#clm-…). The dataset is CC BY 4.0; see the data page. Dataset DOI: 10.5281/zenodo.22739101