「コーヒーが苦いのはカフェインのせい」 (in translation: "coffee is bitter because of caffeine") — probably the most widely believed explanation for bitterness there is. Yet tracing through the literature, this explanation falls short twice over. Caffeine itself is disqualified as 「苦味の代表」 (in translation: "bitterness's representative"), and the leading players in coffee's bitterness are born not in the bean but within the process of roasting.
What this page establishes
Caffeine has been confirmed to act not only through bitterness receptors (TAS2R) but also through multiple pathways such as adenosine receptors and ion channels, and a review states that 「カフェインを苦味の典型例とみなすべきではない」 (in translation: "caffeine should not be regarded as a typical example of bitterness").
Caffeine has been confirmed in cell experiments to activate 5 receptors — hTAS2R7, TAS2R10, TAS2R14, TAS2R43, and TAS2R46.
A considerable part of coffee's bitterness comes from substances born during roasting (quinides, 4-vinylcatechol oligomers, roasting-degradation products of mozambioside), and multiple studies have confirmed that the dominant bitter substance's identity changes between light roast and dark roast.
The bitterness threshold of the compound mozambioside differs by roughly 2-fold between a 2015 report and a 2025 report, even within the same lineage of laboratories, and is not yet settled on a single value.
Caffeine, a complicated substance
Caffeine is certainly bitter — in cultured cell systems it strikes as many as 5 different bitterness receptors at once. But at the same time, because it also acts through pathways that do not run through bitterness receptors — adenosine receptors, intracellular calcium, PDE inhibition, and more — it has been concluded to be 「苦味の普遍的なプロトタイプとしては不適」 (in translation: "not suitable as a universal prototype of bitterness"). The fact that caffeine still tastes bitter to people who lack the TAS2R43 gene (about 18–30%) also shows that a single receptor cannot finish the explanation.
The abstract of an independent functional calcium-imaging study explicitly states, as background, "caffeine, which activates five human bitter taste receptors" (a restatement of the Meyerhof-lineage finding). In that study's own assay as well, TAS2R43/46 responded to caffeine (less sensitively than to mozambioside). This shows that the attribution of 5 receptors was still maintained as a premise in the research community as of 2020.
Limitations: The names of the 5 receptors (TAS2R7/10/14/43/46) are not listed in the abstract. The primary data are Meyerhof 2010 (the related record; full text not reached), and this row remains a restatement plus partial (TAS2R43/46) direct confirmation. The author list includes Behrens and is not independent of the Meyerhof lineage.
Scope: Caffeine's experimental use as a bitter stimulus
Note: The review in question also discusses receptor-independent pathways in addition to the canonical pathway.
Caffeine acts through multiple TAS2R-independent pathways: µM affinity at adenosine receptors A1/A2A/A2B (taste cells express A2B), a rise in intracellular Ca via ryanodine receptors, ITPR3 inhibition, a rise in cGMP via PDE inhibition, and TRPA1 — activated in mice but inhibited in humans (a species difference). The conclusion: "caffeine should not be considered a prototypical bitter taste".
Limitations: A review. States explicitly that knockdown confirmation of action via endogenous T2Rs in taste cells has not been carried out. The receptors are not conserved across species, and the mechanism is species-dependent.
Caffeine (0.1–10 mM) activates mouse TRPA1 (the EC50 is stated in the text as "between 1 and 2.5 mM"), while human TRPA1 "responded to 100 uM AITC, but not to 5 mM caffeine", and suppressed the current under chase application in the presence of AITC. 8 WT mice avoided 5 mM caffeine water (P = 0.012), but 10 TRPA1-KO mice did not avoid it (P = 0.096). Of 63 capsaicin-responsive DRG neurons in WT, 20 responded to both caffeine and AITC, versus 0 of 115 in KO. The abstract closes: "caffeine does not activate human TRPA1; instead, it suppresses its activity". This is primary-level evidence that the receptor pathway for caffeine's response is not conserved across species, supporting the claim that caffeine cannot be treated as a universal bitter prototype.
Limitations: The perceptual consequence of human TRPA1 suppression (how it feels to humans) has not been tested. The authors themselves discuss human caffeine bitterness as proceeding via T2R, so this study demonstrates a species difference in a non-T2R pathway, not a replacement of the whole taste pathway. The authors state that the KO could not be used as a negative control for immunostaining.
caffeine remains bitter even for people lacking TAS2R43supported
Scope: The correspondence between the frequency of the gene deletion in human populations and perception
Note: An example that a single receptor cannot explain bitterness perception. "About 18-30%" is a review-level figure. Note that the primary data (946 people / 52 populations) give a deletion-allele frequency of 0.33, a different kind of quantity from "the proportion of individuals lacking the gene".
About 18-30% of the population lacks the TAS2R43 gene, yet caffeine's bitterness is perceived.
Limitations: Via a review. The primary source for the deletion frequency is not included.
Context: Gives the population-genetic basis underlying this claim from primary data rather than via a review: in 946 subjects from 52 populations worldwide, copy number for both TAS2R43 and TAS2R45 ranged from 0 to 2, and the 43delta and 45delta deletions occurred at high frequency worldwide (0.33 and 0.18). The TAS2R43 deletion spans 37.8 kb, including the entire coding region, and the paper states that in Europe more than 20% of subjects lacked this gene. No signal of natural selection was found, and the authors attribute this pattern to genetic drift.
Limitations: What this study measures is gene copy number, not caffeine perception, so the study itself does not show that caffeine remains bitter in the absence of TAS2R43. It also refines the figure carried over by the review: the reported quantity is a deletion-allele frequency of 0.33, which is not the same as the proportion of people lacking the gene; the paper gives a regional figure for Europe (more than 20%) rather than a single worldwide percentage. The 18-30% range recorded from Poole 2017 should therefore be treated as a review-level approximation.
The chlorogenic acids abundant in green beans are turned, by the heat of roasting, into quinides (chlorogenic acid lactones) — a group of compounds carrying 「コーヒーらしい苦味」 (in translation: "coffee-like bitterness"). Roast the beans deeper still, and the quinides themselves break down, generating 4-vinylcatechol oligomers, a group of compounds with a harsh (stinging) bitterness. That the 「質」 (in translation: "quality") of dark-roast bitterness differs from that of light roast is not a matter of mood — it is chemistry.
Scope: Formed preferentially at light-to-medium roast. At heavy roast it decomposes, generating harsh-bitter 4-vinylcatechol oligomers
Note: The chemical basis for how the bitterness signature of coffee changes with roast degree and extraction conditions.
In extracts made by systematically varying the roast conditions, mono- and di-caffeoyl quinides (coffee-like bitterness) formed preferentially at light-to-medium roast, and decomposed at heavy roast to generate 4-vinylcatechol oligomers (harsh bitterness). This matched the change in bitterness quality found in sensory evaluation.
Limitations: Abstract only checked. Individual compound concentrations and panel details were not confirmed from the full text.
From a dry-heat model of 5-O-caffeoylquinic acid, the source isolated and determined the structure of a previously unknown bitter lactone, 3-O-caffeoyl-epi-gamma-quinide. Its recognition threshold for bitterness was low, at 58 µmol/L, and its presence in coffee beverages was confirmed by LC-MS/MS (MRM). As the 6th bitter quinide, joining the 5 already known caffeoyl quinides, this adds further support to quinides as a class contributing to coffee bitterness.
Limitations: Abstract only checked. The panel and method for the threshold measurement are not stated in the abstract. This is the same research group as Blumberg 2010 (Frank/Blumberg/Hofmann), so it is not an independent-lab replication.
Context: A group independent of the Hofmann laboratory (UFRJ/Vanderbilt) quantified 7 chlorogenic acid lactones (quinides) formed during roasting of 2 arabica cultivars plus robusta, by HPLC comparison against synthetic standards. The most abundant, 3-CQL, reached a maximum of 230±9 mg/100 g (arabica) / 254±4 mg/100 g (robusta) at light medium roast (about 14% weight loss), with maximum CGL formation reaching about 30% of the precursor. Because the 3-CQL/4-CQL ratio reverses relative to the precursor ratio in green beans, the source proposes a pathway in which chlorogenic acid isomerizes before lactonizing during roasting. This independently supports the chemical scope of this claim — that quinides form during roasting and depend on roast degree.
Limitations: Abstract only checked. There is no sensory or bitterness data, and the contribution to bitterness itself was not verified. Decomposition at heavy roast is suggested by the phrase "peak values at light medium roast," but this is not stated explicitly in the abstract.
The 7 chlorogenic acid lactones (CGL/quinides) identified in roasted coffee, individually named: 3-CQL, 4-CQL, 3-pCoQL, 4-pCoQL, 3-FQL, 4-FQL, and 3,4-diCQL. Ranking second by abundance, 4-CQL was 116±3 mg/100 g (dry weight) in C. arabica and 139±2 mg/100 g in C. canephora.
Limitations: Abstract level only (full text not reached). Quantitative values for the 5 compounds other than 3-CQL (pCoQL, FQL, diCQL) are not in the abstract. Sensory or bitterness data themselves are not included.
Scope: Formed under heavy-roast conditions alongside the decomposition of quinides, producing a harsh bitterness quality
Note: The chemical substance behind how the “quality” of bitterness changes with roast degree.
Under heavy-roast conditions, quinides decompose and generate harsh-bitter 4-vinylcatechol oligomers (compounds 12–20). This matches the change in bitterness quality observed sensorially.
A dry-heat model of caffeic acid produced a strong bitterness reminiscent of that of "strongly roasted espresso-type coffee." Using taste dilution analysis to narrow down the bitter compounds, the source identified a total of 10 compounds — 1,3-bis(3',4'-dihydroxyphenyl)butane, the corresponding butene, and 8 multiply hydroxylated phenylindanes (recognition thresholds 23–178 µmol/L). Each compound was also confirmed in coffee beverages by LC-MS/MS (MRM), and the structures strongly suggest formation via oligomerization of 4-vinylcatechol released from the caffeic acid moiety during roasting.
Limitations: Abstract only checked. The threshold range of 23–178 µmol/L applies to the compound set as a whole; individual assignments are not in the abstract. Because this is the same Hofmann laboratory as Blumberg 2010, it is not a fully independent replication (a different experimental system from the same group).
Context: A Japanese group independent of the Hofmann lineage (Osaka City University and Ajinomoto AGF) used a high-temperature processing model of caffeic acid (simulating food roasting) and determined the structures of various oligomers as heat-reaction products that were "produced via the intermediate 4-vinylcatechol." Coupling products of caffeic acid and 4-vinylcatechol formed mainly at 140-170 °C, while at 200 °C further reaction products formed via addition and elimination of catechol units predominated. This independent laboratory confirmed the formation pathway assumed by the claim (caffeic acid → 4-vinylcatechol → oligomers) and its temperature dependence.
Limitations: Abstract only checked. The study's purpose was xanthine oxidase inhibitory activity; there is no bitterness or sensory data. Quantification in real coffee samples is also not included in this study. An independent replication of the sensory aspect — the harsh bitterness quality — still does not exist.
Another leading player is mozambioside, a glycoside found only in the arabica species. The group of compounds generated when it breaks down under roasting produces bitterness at a lower concentration than the parent compound, and moreover the receptors it strikes are TAS2R43 and TAS2R46 — the very same 2 that, in the previous note, were shown to receive chicory's bitterness. At the receptor level, coffee and chicory turned out to be neighbors.
mozambioside degrades into more potently bitter compounds via Roasting-derived breakdown products (glucosylated cafestol derivatives, bengalensol, and others) have a lower bitterness threshold than the parent compound (27–80 µM vs 132 µM) and activate TAS2R43/46supported
Scope: HEK293T cell assay plus human threshold measurement. A mixture at coffee-level concentrations was recognized as bitter by 80% of the panel. Sensitivity correlated with whether the TAS2R43 gene locus was intact
Note: That chicory bitterness is received by the same TAS2R43/46 connects to Research Note 002.
The structures of 5 heat-degradation products were determined by NMR/UHPLC-ToF-MS. Of these, 3 had a lower receptor-activation threshold than the parent compound. Docking analysis examined the binding mode to TAS2R43/46. The compounds were also detected in actual roasted-coffee extracts.
Limitations: Abstract only checked. This is a cultured-cell system, not human perception itself.
21.0–170.4 nmol/g in ground coffee, with an extraction rate of about 41–128%. The human threshold of the main breakdown products was 27–80 µM (parent compound 132 µM). A mixture of 1–8 at coffee-level concentrations was recognized as bitter by 80% of the panel. Sensitivity correlated with an intact TAS2R43 gene locus.
Limitations: Abstract only checked. Panel size is unknown. This is only at the stage of 「寄与を示唆」 (in translation: "suggesting a contribution"), not a quantification of the contribution rate.
In functional calcium imaging, TAS2R43/46 responded to mozambioside with far higher sensitivity than to caffeine. The structurally related compounds bengalensol, cafestol, and kahweol also activated the same receptor pair more strongly than caffeine did. Kahweol is a weak partial agonist of TAS2R43 and showed a suppressive effect when co-applied with mozambioside. This lends prior support to the receptor assignment reported by Czech 2024 and Bichlmaier 2025, that 「モザンビオシド系化合物の受け手は TAS2R43/46」 (in translation: "the receptors for mozambioside-class compounds are TAS2R43/46").
Limitations: Abstract only checked. This is receptor data for the parent compound and structurally related compounds, not the roasting breakdown products themselves (glucosylated cafestol derivatives, etc.). This is research from the Leibniz/TUM cluster with personnel overlap with Czech 2024 and Bichlmaier 2025, so it is not a fully independent-lab replication.
The formal names of the 5 mozambioside heat-degradation products: 17-O-β-D-glucosyl-11-hydroxycafestol-2-one, 11-O-β-D-glucosyl-16-deoxycafestol-2-one, 11-O-β-D-glucosyl-(S)-16-deoxy-17-oxocafestol-2-one, 11-O-β-D-glucosyl-15,16-dehydrocafestol-2-one, and 11-O-β-D-glucosyl-(R)-16-deoxy-17-oxocafestol-2-one. Of these, the (S)-form, the 15,16-dehydro form, and the (R)-form (3 compounds) showed a lower receptor-activation threshold than the parent compound, and 2 of these 3 (the 15,16-dehydro form and the (R)-form) were also detected by UHPLC-ToF-MS in actual roasted-coffee extracts.
Limitations: This information was extracted from the structured abstract (JSON-LD metadata); the Materials and Methods/Results body of the full text was not reached. This is a cultured-cell assay, not human perception itself.
Show datacompound-mozambioside —degrades_into_more_potent_bitter_compounds_via→ "Roasting-derived breakdown products (glucosylated cafestol derivatives, bengalensol, and others) have a lower bitterness threshold than the parent compound (27–80 µM vs 132 µM) and activate TAS2R43/46"
For this mozambioside, the human bitterness recognition threshold was 60±10 µM in the 2015 report and 132 µM in the 2025 report — a roughly 2-fold discrepancy even within the same lineage of laboratories. NIGAMI LAB keeps both, preserved under the 「係争中」 (in translation: "contested") label rather than choosing one as the correct value. That is because a bitterness number is a function of the panel and the method.
mozambioside — bitter recognition threshold: 60±10 µM (2015) and 132 µM (2025) — roughly a 2-fold discrepancy within the same lineage of laboratoriescontested
Scope: Human recognition threshold in aqueous solution
Note: NIGAMI LAB's first contested claim. The threshold is panel- and method-dependent, an example that it cannot be treated as a single fixed value.
Mozambioside's bitterness recognition threshold is 60±10 µmol/L.
Contradicts: Reports the threshold of the same compound as 132 µM — about 2.2 times the 60±10 µM reported in 2015. Panel, purity, and methodological differences are candidate causes.
Limitations: Only the abstract was checked. The cause of the discrepancy cannot be identified from either abstract alone.
Show datacompound-mozambioside —has_bitter_recognition_threshold→ "60±10 µM (2015) and 132 µM (2025) — roughly a 2-fold discrepancy within the same lineage of laboratories"
The amount of bitterness-receptor mRNA on the tongue, and habitual caffeine intake — together, these two factors are reported to have explained 47% of the variance in caffeine bitterness ratings (a small study with n=20). Even from the same cup, a different bitterness may be rising up for you than for the person sitting next to you.
caffeine — perceived bitterness is associated with: Habitual caffeine intake and bitter-receptor mRNA abundance in taste papillaesupported
Scope: A small-scale human study reported in the cited paper
Note: The association does not establish the direction of causation, nor can it be generalized to every population.
n=20. gLMS bitterness ratings of 8 mM caffeine were associated with combined TAS2R10/43/46 mRNA abundance (r(18)=0.47) and with habitual intake (t(18)=2.78, p=.012), with the 2 factors together explaining 47% of the variance in ratings. TAS2R43 alone also showed r(18)=0.54, p=.014. No comparable association was seen for quinine.
Limitations: A cross-sectional study of n=20. Receptor genotypes other than TAS2R38 were not controlled for. The direction of the association (whether habitual drinkers feel it more or less strongly) is not stated explicitly in the text, and the discussion notes that prior intake-manipulation studies have reported inconsistent results.
Context: In large-scale genetic epidemiology too, people with higher genetic sensitivity to caffeine bitterness drink more coffee (+0.146 cups/day per SD) — consistent with the direction of 「常飲と感受性の関連」 (in translation: "the association between habitual drinking and sensitivity"), and interpreted as learned positive reinforcement.
Limitations: Intake is self-reported. The causal path from sensitivity to intake is not a direct measurement of a learning mechanism.
Summary: from 「カフェインの苦味」 (in translation: "caffeine's bitterness") to 「焙煎の苦味」 (in translation: "roasting's bitterness")
Coffee's bitterness turned out to be, not the story of a single molecule called caffeine, but the story of a group of compounds that the process of roasting keeps rewriting.
Scope: Multiple lines of evidence: human sensory testing, receptor assays, and quantitative chemistry
Note: A group of counterexamples to the common belief that 「コーヒーの苦味＝カフェイン」 (in translation: "coffee's bitterness equals caffeine"). The quantitative contribution ratio is not recorded, since the full text was not confirmed.
Roasting-generated compounds other than caffeine (quinides, 4VC oligomers) define the bitterness quality, and the bitterness signature can be controlled through roasting/extraction conditions.
The primary literature on quinides and mozambioside-related compounds is, at this point, at the abstract-checked stage (see each card's verification-status display). 2 caffeine papers (Poole 2017, Lipchock 2017) have been checked in full text.
A quantification of the contribution ratio each compound group holds within coffee's overall bitterness is not written into this note, since the full text was not confirmed.
The receptor findings are measurements in cultured cell systems.
Frequently asked
Is coffee's bitterness the taste of caffeine?
Caffeine has indeed been confirmed to activate some bitterness receptors, but it has also been confirmed to act through multiple pathways that do not go through bitterness receptors, such as adenosine receptors and ion channels. For this reason, a review states that caffeine should not be treated as a typical example of bitterness. In fact, people who lack the gene for the receptor TAS2R43 make up roughly 18–30% of the population, but caffeine's bitterness is reported to be perceived by such people too, so an explanation relying on a single receptor alone is not enough. Multiple studies show that much of coffee's bitterness comes from substances other than caffeine, such as quinides and 4-vinylcatechol oligomers born during roasting.
Does coffee get more bitter the deeper it is roasted?
From light roast to medium roast, it has been confirmed that quinides, which carry a coffee-like bitterness, are preferentially generated. It has also been confirmed, by both sensory evaluation and chemical analysis, that when roasting is intensified, these quinides break down and 4-vinylcatechol oligomers, which carry a harsh (rough) bitterness, are generated instead. In other words, it is known that as roasting deepens, it is less the amount of bitterness and more the quality of bitterness that changes, but how far the amount itself increases monotonically has not been confirmed within the scope of this material.
What is not known
Why mozambioside's bitterness threshold differs by roughly 2-fold between studies (which of panel, purity, or methodological differences is at work) has not yet been confirmed.
The association between habitual caffeine intake, receptor mRNA levels, and bitterness ratings remains confined to a single cross-sectional study of about 20 people, and whether sensitivity determines intake or intake changes sensitivity — that direction has not yet been confirmed.
It is known that how caffeine modifies human TRPA1 works differently depending on the species, as with mice, but the perceptual consequence itself in humans has not been sufficiently tested.
No material quantifying how much roasting-degradation products' receptor activation contributes to coffee's overall bitterness, as a proportion, could be found.