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.
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 coefficients between a single peptide's concentration and the bitterness score stayed at 0.56 and 0.58, versus r=0.84 for a combined 2-peptide index (the sum of each concentration divided by its highest value in the data set) (see limits).
A total of 16 bitter peptides were identified from the water-soluble fraction of aged Gouda. All derive from casein proteolysis: 11 are previously unreported β-casein-derived peptides (including 6 derived from β-CN(57-69) and 2 from β-CN(198-206), among others), and 6 are derived from αs1-casein.
Limitations: The abstract does not give sample details such as the cheese's ripening period or origin.
As the compounds responsible for the bitterness of aged sharp Cheddar, the 5 peptides with the highest bitterness intensity (all derived from β-casein) were identified by de novo sequencing.
Limitations: A corrigendum was issued in 2019 (contents not checked). The abstract gives no absolute values for concentration or threshold.
Limits: The correlation coefficients between individual peptide concentration and bitterness score stayed at 0.56 for αs1-CN f1-9 and 0.58 for β-CN f193-209. An index that adds together, for each cheese, the 2 ratios of each peptide's concentration to its highest value in the data set gave r=0.84 (Fig. 3). The authors state that other peptides certainly also contribute to bitterness.
Limitations: A limit on claims that attribute cheese bitterness to a single peptide. The regression is from a reduced-fat model system.
bitter peptides — recognition thresholds: 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.supported
Scope: Peptides purified from aged Gouda cheese and then synthesized.
Note: The abstract gives neither the threshold-measurement method nor the panel size.
In sensory evaluation of the purified and synthesized peptides, 12 showed clear bitterness, with recognition thresholds of 0.05-6.0 mmol/L. The decapeptide YPFPGPIHNS had a caffeine-like bitterness quality and the lowest threshold, 0.05 mmol/L.
Limitations: The abstract gives neither the panel size nor the threshold-measurement procedure.
Context: Verbatim: 'the synthetic peptides exhibited a bitter taste with threshold values 0.25 and 0.16mM'. The 2 synthesized hydrophobic peptides correspond to bovine β-casein f61-67 (H-Pro-Phe-Pro-Gly-Pro-Ile-Pro-OH) and f60-66, chosen as a common portion of bitter peptides isolated from cheese. In this way, an independent group also places the threshold of casein-derived cheese bitter peptides in the same sub-mM-to-mM range as the Gouda set (0.05-6.0 mmol/L).
Limitations: Because this is a different peptide from the Gouda set (a middle portion of β-casein rather than the C-terminal Gouda-derived decapeptide), what it corroborates is the order of magnitude, not the individual value. The abstract does not give the panel size or the threshold procedure.
Limits: The taste of the synthetic peptides was judged by a panel of 4 people, and the threshold was reported as a ratio relative to caffeine as the reference substance (Rcaf). As the peptide was shortened from the full-length form, the threshold concentration rose monotonically (bitterness weakened).
Limitations: The panel size is small, and this is not an absolute threshold.
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).
During ripening, GPVRGPFPIIV, YQEPVLGPVRGPFPI, and MPFPKYPVEP increased 28.7-fold, 3.1-fold, and 1.8-fold respectively; APHGKEMPFPKYPVEPF was detected only in sharp Cheddar; and MAPKHKEMPFPKYPVEPF was unchanged. A mild Cheddar model with the 5 peptides added at matched concentrations showed no significant difference in bitterness intensity from real sharp Cheddar. The main contributor was GPVRGPFPIIV.
Limitations: The abstract does not give absolute concentration values or panel details.
In aged Cheddar, the concentration of the β-casein C-terminal bitter peptide β-CN f193-209, quantified by MALDI-TOF, correlated with the bitterness intensity rated by a trained panel: the correlation coefficient was 0.803 at 180 days and 0.554 at 270 days. That is, an independent group has quantitatively linked the β-casein C-terminal peptide to Cheddar bitterness, which is consistent with the results of the reconstitution experiment in which β-casein-derived peptides (including the f193-209 fragment) reconstituted the bitterness of sharp Cheddar.
Limitations: This is only a correlation, and only a single peptide was tracked; the authors attribute the drop in correlation at 270 days to other bitter peptides or masking compounds, which itself constrains attribution to a single peptide. The abstract gives no panel details.
Note: 2 independent lines of cell assay point to the same receptor. The contribution in vivo is not established.
The bitter dipeptides Gly-Phe and Gly-Leu (25 mM) activate hTAS2R1 most strongly, with no or weak response at other receptors. The tasteless dipeptide Gly-Gly did not activate any hTAS2R (tested up to 100 mM). hTAS2R1 was more sensitive to bitter peptides than to other bitter compounds such as denatonium, picrotin, salicin, and caffeine.
Limitations: The authors state explicitly that background signal also rises at higher concentrations. This is a cell-assay result; the receptor's contribution in human sensory perception is not directly shown.
Bitter tripeptides activate T2R1 more strongly than dipeptides, with the strongest, Phe-Phe-Phe (FFF), having an EC50 in the micromolar range. Homology modeling places the bitter peptides binding to the same pocket on the receptor's extracellular face (a cap structure formed by TM1/2/3/7 and ECL1/2).
Limitations: The actual EC50 value is not given in the abstract. The modeling is not an experimental structure.
Statements: "TAS2R1 exhibited strong responses to the two bitter peptides GL and GF but not to the nonbitter GG" and "The receptor that was found to be activated by the highest number of peptides is the receptor TAS2R1." In Table 1 (List of Amino Acids/Peptides Activating Human Bitter Taste Receptors), the TAS2R1 row lists "F, GF, GL, IF, LW, FI, FL, WW, GLL, IQW, LKP, FFF, WWW, YPFPGPIHNS, LVYPFPGPIHN".
Limitations: Includes data integrated from prior literature (including Maehashi 2008 and Upadhyaya 2010); within the scope of extraction in this review, it has not been possible to fully separate out which data are this paper's own new experimental results.
The 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).
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.supported
Scope: A review on cheese bitterness (1996). It covers isolation and detection methods for bitter peptides, their physicochemical properties, and control methods.
The review states that bitterness is an inevitable consequence of proteolysis, and reviews the relationship between proteolytic activity and bitter-peptide accumulation, along with isolation and detection methods and physicochemical properties.
Limitations: This is a 1996 review and does not reflect later molecular-biology findings (for example, PrtP specificity).
Among the peptides that have been linked to aged Cheddar's bitterness is β-casein-derived f193-209. This peptide is generated by chymosin. Even when the starter strain differed, its concentration stayed within a range of 0.3 to 1.0 mg/g, and no significant difference by starter was detected. 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.
bitter peptides includes a chymosin-derived member: β-CN f193-209: a chymosin-derived peptide that has been associated with bitterness in cheese. Its concentration was 0.3-1.0 mg/g; the effect of ripening time was significant, but no difference among starters was detected (P>0.5). A combined index with αs1-CN f1-9 has been proposed for predicting the tendency toward bitterness (r=0.84).supported
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.
The authors build on earlier literature (refs 4 and 28) associating αS1-CN (f 1-9) and the chymosin-derived peptide β-CN (f 193-209) with bitterness in cheese. The concentration of β-CN f193-209 was 0.3-1.0 mg/g (Table 4, all sampling times); the effect of ripening time was significant (P<0.0001), and no effect of starter was detected (ANOVA P>0.5). An index combining both peptides is proposed as useful for predicting the tendency toward bitterness (r=0.84).
Limitations: This paper itself contains no threshold measurement of β-CN f193-209's bitterness; the association depends on cited references (refs 4 and 28).
The concentration of β-CN f193-209 measured by MALDI-TOF in aged Cheddar correlates with sensory bitterness intensity (r = 0.803 at 180 days, 0.554 at 270 days), supporting the use of this peptide as a quantifiable marker linked to cheese bitterness.
Limitations: Because the correlation weakens the longer ripening continues (0.554 at 270 days), this peptide is a partial marker rather than the sole driver. Extraction at the abstract level.
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.
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).
Cells expressing hTAS2R1, hTAS2R4, hTAS2R14, and hTAS2R16 respond to bitter fraction M (2.5 mg/mL; moderately bitter, equivalent to the bitterness of 0.125 mM quinine), obtained by gel filtration of a 12-hour trypsin digest of casein (1%), with hTAS2R1 and hTAS2R4 responding especially strongly.
Limitations: The fraction is a mixture; the response is attributable only at the level of the fraction.
Limits: The authors state explicitly that the bitter fraction contains numerous structurally different peptides and is likely to include non-bitter peptides as well. In addition, background signal rises at high peptide concentrations, and hTAS2R16 showed no significant response to the synthetic dipeptides (though it did respond to its specific ligand, salicin).
Limitations: The correspondence between individual peptides and individual hTAS2Rs cannot be established with this experimental system.
This 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).
Table 2: threshold (uM) for TAS2R1: oligoporin A=30, oligoporin B=30, oligoporin D=30, oligoporin E=30, oligoporin F=(blank, not activated), infractopicrin=(blank). The body text states (regarding oligoporin D) that activation of receptor-transfected cells was seen for TAS2R1, TAS2R4, TAS2R14, and TAS2R46, and more weakly for TAS2R41.
Limitations: This is a result for a compound isolated from a single mushroom species; no sensory evaluation as a food has been carried out.
Whether 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.
cheese — bitterness develops depending on: Specificity groups of Lactococcus lactis lactocepin (an extracellular proteinase). At 6 months, bitterness was 1.8/9 for proteinase-negative-strain cheese against 3.5, 4.2 and 4.7 for groups a/e/h. Starter and ripening time were each significant (P<0.0001); pooled over ripening time all 3 strains were significantly more bitter than the negative strain, but at 6 months alone the difference between group a and the negative strain was not significant. Group h was the only strain significantly more bitter than the negative strain at every sampling time.supported
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.
Bitterness scores (0-9 scale) at 6 months: proteinase-negative PH=1.8, PHa=3.5, PHe=4.2, PHh=4.7. Starter and ripening time were each significant (P<0.0001), and in the strain means pooled over ripening time (PH 1.5, PHa 3.1, PHe 3.2, PHh 3.9) all 3 lactocepin-carrying strains were significantly more bitter than PH. At 6 months alone, the difference between PHa and PH was not significant, and PHh was the only strain significantly more bitter than PH at every sampling time. αs1-CN f1-9 accumulation ran in parallel: 12.1 mg/g for PHh versus 2.9 mg/g for PH (at 6 months).
Limitations: A reduced-fat, single-strain model system. Even the authors limit extrapolation to commercial mixed starters to a suggestion based on the gene-exchange result. The definitions of "0" and "1" on the scale differ between the methods section (0 = not bitter, 1 = just perceivable) and the Table 5 footnote (1 = no perceptible bitterness).
Context: A follow-up genetic-engineering study: through a 2-step gene replacement, the substrate-binding determinants of prtP, which encodes the group h (bitter-type) lactocepin, were converted to the group b (nonbitter-type) variant in 2 industrial L. lactis strains. RP-HPLC of lactocepin activity against αs1-casein f1-23 confirmed the change in specificity, and there was no significant change in the milk's acid production or autolysis. The framework treating group h as the bitter type and group b as the nonbitter type puts the link between specificity group and bitterness into a form usable for strain improvement.
Limitations: This is the same research group as the 2002 cheese study, not an independent replication. The abstract reports only the conversion of enzyme specificity; it does not present cheesemaking or sensory bitterness data.
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.
bitter peptides is produced through: Casein breakdown by PrtP-positive Lactococcus lactis strains. When LTH 7122 was grown in reconstituted casein medium, PrtP peptidase activity rose in the growth phase (5 hours) under higher protein and calcium (8.5% protein, 70 mM calcium), and casein-derived bitter peptide concentrations were also higher. Transcription of the prtP gene did not clearly rise in the high-protein medium. In tests in milk citrate broth (MCB), according to the abstract, adding 50 mM CaCl2 lowered transcription by more than 50% in both strains, while activity fell 15% in LTH 7123 and rose slightly in LTH 7122. The change in activity and the change in transcription level did not necessarily agree.supported
Scope: Culture experiments in milk citrate broth (LTH 7122 and 7123, 3 and 5 hours) and in reconstituted casein medium (protein 3.4% and 8.5% w/v, LTH 7122 only, 5 and 24 hours). An experiment separate from the cheese sensory test.
Note: At low calcium concentration (+5 mM in MCB), both prtP transcription and activity rose, but at high concentration (+50 mM) the responses of transcription and activity diverge — reported as a nonlinear relationship.
In tests in milk citrate broth (MCB) (LTH 7122 and 7123, 3 and 5 hours), adding 5 mM CaCl2 raised both prtP transcription and PrtP activity, while adding 50 mM lowered transcription. The abstract states that at 50 mM transcription fell by more than 50% and that activity fell 15% in LTH 7123 and rose slightly in LTH 7122 (the body text describes the LTH 7123 activity as comparable to the basic medium). In tests in reconstituted casein medium (LTH 7122 only, 5 and 24 hours), PrtP activity rose at 5 hours at 8.5% protein and 70 mM calcium (at 24 hours it was similar to 3.4% protein), and bitter peptide concentrations were also higher under high-protein, high-calcium conditions. Transcription did not clearly rise in the high-protein medium.
Limitations: The relationship between transcription level and peptidase activity is not consistent across strains, media and time points. The percentage changes at 50 mM are the abstract's figures, and the figure values were not checked. Bitter peptide concentrations were measured in the culture medium, in an experiment separate from the cheese sensory test.
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.
cheese — bitterness develops 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.supported
Scope: Experimental fresh cheese (CoF) made by fermenting high-protein concentrated milk reconstituted from micellar casein powder (8.5% protein, about 74 mM calcium). Bitterness rated by 12-13 members per session of a 25-member trained 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), and the discussion states that cheese made with LTH 7112 was significantly more bitter than the directly acidified control.
CoF fresh cheese fermented with the PrtP-positive strains LTH 7123 and LTH 7122 scored bitterness 3.9 and 4.1 respectively; the PrtP-negative strains LTH 7117 and LTH 7112 scored 1.7 and 2.8 (0-5 scale). The text states that cheese made with the positive strains was significantly more bitter than with the negative strains (p<0.05; the significance letters for each group are in Figure 1, which was not checked).
Limitations: There was no significant difference between the non-fermented, directly acidified control (CoF control) and the commercial reference product (FCo). The significance letters for each group are only in Figure 1, which was not checked.
Limits: The conclusion section summarizes that "the prtP-negative strains did not introduce bitterness," but in the results section, cheese made with the PrtP-negative strains (1.7 for LTH 7117, 2.8 for LTH 7112) was also more bitter than the commercial fermented reference FCo (0.5), and LTH 7117 was rated "slightly bitter". The discussion states explicitly that cheese made with LTH 7112 was significantly more bitter than the non-fermented, directly acidified control CoF (1.0), and that even negative strains may produce some bitterness. There is a mismatch in the strength of the wording between the summary and the results and discussion.
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.
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.supported
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.
Reduced-fat Cheddar cheese made with an integrated culture system using SK11 as the primary acid-producing strain had substantially lower bitterness intensity than cheese made with a commercial Cheddar starter. Diacetylactis JVI significantly increased the concentrations written in the abstract as 'diacetylacetoin and acetate', and the sensory results showed increased buttery (diacetyl) flavor.
Limitations: The abstract does not give a specific figure for the bitterness score difference. Nor does it report a statistical test of the bitterness difference, and the result does not mean the cheese was free of bitterness.
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.
cheese — bitterness develops 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.supported
Scope: Aged Cheddar cheese compared across 4 groups: control, D. hansenii alone, Y. lipolytica alone, and both species combined (the country of the experiment is not given in the abstract).
Note: The specific sensory scale and panel size are not stated in the abstract.
Aged Cheddar cheese to which D. hansenii or Y. lipolytica was added alone as an adjunct showed accelerated development of Cheddar flavor, but this was accompanied by the occurrence of bitter flavor as well.
Limitations: The size and scale of the sensory panel cannot be determined from the abstract.
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.
cheese — bitterness develops depending on: For Cheddar made with the nisin Z-producing strain Lactococcus lactis subsp. lactis biovar diacetylactis UL719, the abstract says bitterness was stronger than in the control, but the sensory results in the text say it was weaker than in cheese made with the same starter without nisin, so the two conflict. The strain increased peptide formation in the cheese water extract.preliminary
Scope: Cheddar cheese, aged 6 months. Sensory evaluation was carried out at the 6-month point.
Note: 2026-10-21 (full-text re-evaluation, batch 2): in the full text, the abstract (bitterness enhanced) and the sensory results in the text (weaker than without nisin) run in opposite directions; as neither can be settled, the claim was lowered from supported to preliminary.
The abstract says Cheddar made with the nisin Z-producing strain UL719 was more acidic and bitter than the control, whereas the sensory results in the text say this cheese was less bitter than cheese made with the same starter without nisin, so the two run in opposite directions. Hydrophobic peptide peak areas at 214 nm at 6 months (×10^7; not sensory values) were ISN 8.21, NF 8.80 and COM 8.04, with no significant difference.
Limitations: The conflict between abstract and text cannot be resolved within the source (what the abstract compared against cannot be confirmed). Sensory evaluation was a single session at 6 months.
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.
cheese can be debittered by: Cheddar with an adjunct culture of Lactobacillus casei-casei L2A was less bitter and ranked higher in flavour quality (the text attributes this mainly to Lb. casei); an added sensory benefit from combining it with the nisin-producing strain or liposome-encapsulated nisin Z cannot be established.preliminary
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. NFB, which ranked first for overall quality in the text, had Lb. casei added alone without nisin, and ISNB, combined with the nisin-producing strain, ranked below NF. (2026-10-21, expansion stage N5 stage C batch 5: Codex's check against the source confirmed that the sensory test was an overall-quality rating by 5 people with no numerical bitterness intensity and no test, and that milk heat treatment also differed with and without Lb. casei; lowered from supported to preliminary.)
Cheese with an adjunct Lb. casei was less bitter and ranked higher in flavour quality. In the overall-quality ranking, NFB (Lb. casei only, no nisin) and LNB (Lb. casei with liposomal nisin) tied for first, followed by NF > ISNB (Lb. casei with the nisin-producing strain) > ISN > COM, with no flavour difference detected between NFB and LNB. The text attributes the debittering mainly to Lb. casei. Hydrophobic peptide peak areas at 280 nm at 6 months (×10^7) were ISNB 0.86, NFB 1.07 and LNB 1.10, and the authors discuss that combining a nisin-producing strain with debittering strains may improve flavour quality.
Limitations: The sensory test ranked flavour quality, not intensity. The panel was 5 people, with no numerical table of bitterness intensity and no test. An added sensory benefit from nisin cannot be established (chemical changes are reported). Milk heat treatment also differed with and without Lb. casei (72 °C vs 80 °C).
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.
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.supported
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.
As the degree of hydrolysis increased from 7.5% to 28.0%, the bitterness sensory score fell from 5.0 to 0.3-2.7 (depending on the protease used, with FU the most effective). The analysis found that a protease cleaving Pro, Leu, Phe, and Val is desirable for effective debittering.
Limitations: The source reports that the bitterness of the whole cannot be predicted by a simple sum of individual peptides' bitterness.
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.
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.supported
Scope: A review on the properties and applications of microbial aminopeptidases.
The review states that microbial aminopeptidases selectively cleave hydrophobic amino acids from the N-terminus and are applied to reduce the bitterness of protein hydrolysates, cheese, and the like.
Limitations: This abstract contains no quantitative comparison of effect by individual strain or enzyme.
「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.
Ney 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).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.
Context: The source explains that this correlation is known as the Q rule, which predicts bitterness for peptides whose Q value is higher than 1400 cal/mol and whose molecular weight is between 100 and 6000 kDa.
Limitations: The original text prints the unit as "6000 kDa"; in light of the usual statement of Ney's rule (molecular weight up to 6000), this appears to be a typo for Da. The original source was not read in this survey.
Contradicts: The source concludes that hydrophobicity data based on Q values do not support Ney's Q rule as a predictor of bitterness for soy peptides. Bitterness showed a nonmonotonic dependence on molecular weight, peaking at a mean molecular weight of about 4000 Da (the first hydrolysate) and 2000 Da (the second), and the fraction below 1000 Da had the weakest bitterness in both hydrolysates.
Limitations: This counterevidence is limited to soy peptides. The abstract does not give the actual Q-value figures or panel size.
The authors state that the result — that the hydrophobic dipeptides Gly-Phe and Gly-Leu are bitter and activate hTAS2R1, while the less hydrophobic Gly-Gly is tasteless and inactive — is consistent, at the dipeptide level, with the hypothesis linking hydrophobicity to bitterness (the Q-value rule).
Limitations: This is consistency across only 3 dipeptides, not a test of the rule's generality.
As a method for predicting whether a peptide is bitter from its amino acid composition, the value Q is calculated from the solubility data of each amino acid; a peptide with Q < 1300 is not bitter, and a peptide with Q > 1400 shows bitterness, the source states.
Limitations: Only the summary was checked. The summary does not specify how the intermediate band of 1300-1400 is treated.
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.
Ney 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.contextual
Scope: Layer: literature. A statement in a 2024 review that cites the original Ney 1971 source.
"In 1971, Ney et al. proposed the Q rule, primarily used to describe and predict the bitterness level of peptides by calculating the frequency of hydrophobic amino acid residues to obtain a Q value. Peptides with a Q value less than 1300 are not bitter, while those with a Q value greater than 1400 are bitter."
Limitations: The original Ney 1971 source (a German-language paper) was not checked by this team; this relies on the summary given in the 2024 review.
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.
Ney 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.preliminary
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).
"Fu et al. used the self-digestion product of shrimp heads as the research model...they found that the core bitter components in the hydrolysis products were mainly concentrated in the peptide segments of 3-5 kDa. Among the different molecular weight components separated, the components with a lower average hydrophobic value actually showed stronger bitterness. Moreover, the positioning of the hydrophobic amino acid residues at the N-terminal or C-terminal of the peptide chain determines their bitter properties, and the bitter response caused by the C-terminal residues is significantly stronger than that of the N-terminal residues."
Limitations: The original (Fu et al.) was not checked by this team; this relies on the summary given by the 2026 review (secondary citation).
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.