3.1 Coverage of the database
The four streams yielded 78 records, merged into 52 single molecules plus 6 compound-class or mixture entries kept separate. The distribution across tiers with human data is the central result of this report:
The four-tier chain is complete for 5 molecules: carvone, menthol, linalool, trimethylamine and vanillin. Eleven molecules have receptor and genetic-variant data but no human neural or emotional data at all. Eight molecules appear in the literature without human data at any of the four tiers — they are in the database because they are cited frequently, not because there is evidence.
What stands out is the pattern of the distribution. The molecules with the most neural data are androstadienone, CO₂ and 2-phenylethanol — none of them an important commercial fragrance material. Conversely, for linalool, limonene, eugenol and carvone we have not found a single fMRI or PET study reporting their individual effects. The gap between media intensity and evidence intensity is largest precisely for the most popular terpenes.
3.2 Tier R — molecule to human receptor
We obtained 76 receptor–ligand pairs across 50 receptor groups, of which 46 groups have data on the human receptor. Distribution by quality of evidence: 5 pairs have cryo-EM structures, 16 pairs have quantitative EC50 values, 43 pairs have been deorphanised but lack an EC50, 6 receptors are candidates never tested with any ligand, and 5 associations have been reported but could not be reproduced.
Three figures gauge the state of this field:
- About 88% of human odorant receptor sequences still have no identified ligand. Progress over twenty years: about 1.6% in 2007, 10% in 2015, 12% in 2024.
- About 17% of deorphanisation results fail to replicate when the assay system is changed. A concrete example: OR1A1 with (R)-carvone was reported with an EC50 of about 150 nM in one study and 126 µM in another — a discrepancy of about 800-fold.
- In the largest database of receptor–odorant pairs, 94% of the more than 51 thousand pairs tested are negative results. This is a field in which most of what has been done is exclusion, not confirmation.
Two principles have direct consequences for compounding. First, one molecule uses different receptors at different concentrations: caproic acid acts via OR2W1 at low concentration and OR1A1 at high concentration; linalool via OR1C1 at low concentration and OR2A25 at high concentration. Dilution does not just weaken an odour; it can switch it to a different sensory channel. Second, binding in vitro does not imply a perceptual consequence: vanillin activates OR10G4 in the test tube, yet OR10G4 genotype has no effect on how people perceive vanillin, while it explains 15.4% of the variance in the intensity of guaiacol. The system has receptor redundancy.
3.3 Tier G — genetic variants and population differences
This is the tier with the clearest commercial consequences, because it quantifies the extent to which the same formula gives different experiences in different groups of people.
The strongest case is β-ionone, the violet floral note widely used in perfumery. The N183D variant of the OR5A1 receptor classifies people as sensitive or less sensitive with 96.6% accuracy across 206 people — close to a Mendelian trait. The less-sensitive allele has a frequency of about 70.4% at the East Asian ancestral locus versus 50.0% in Africa, leading to an estimate that about 49% of East Asians are less sensitive to this compound, versus about 19% in populations of African origin.
The second case is trans-anethole (anise, liquorice). The p.Lys233Asn variant of OR6C70 has a frequency of 57% in East Asia versus 19% in Europe, and it raises both intensity and pleasantness.
The third case reverses intuition about the relationship between intensity and emotion. The loss-of-function allele of OR10G4 makes guaiacol more pleasant by 3.33 to 3.71 points per allele copy, explaining 10.3% of variance across 308 people. And for isoeugenol, the same variant changes pleasantness without changing intensity — the only case of this kind among ten validated pairs. Losing a receptor does not just weaken an odour; it can make the odour more pleasant.
The fourth case shows that differences do not always run in the same direction. The S95P variant of TAAR5, the receptor for trimethylamine (fishy odour), reduces perception of this odour; but its frequency in East Asia is only 0.0054% versus 1.32% in non-Finnish Europeans — about 240 times lower. For this compound, the variant contributes almost nothing to perceptual differences in East Asian populations.
The limits of this tier need to be stated clearly. Across all the data, receptor genotype plus ancestry, sex and age explain a median of only 4.77% of variance in odour perception. Genetics explains a few special cases very strongly and the rest very weakly.
The data gap is systematic: there is no olfactory genetics study in East Asian or Southeast Asian populations. Only three variants have clear frequency figures for East Asia, and all three come from studies that took frequencies from genome databases rather than from perceptual measurements in East Asian people.
3.4 Tier N — the neural code in humans
We obtained 31 molecules named in the Methods section of at least one human neural study. But only 8 molecules meet the full attribution standard, meaning that the single molecule was presented on its own and the results were reported separately for that compound: androstadienone, CO₂, 2-phenylethanol, valeric acid, citral, cinnamaldehyde, 1-butanol and indole. And only a single molecule has a dose curve in the entire human olfactory imaging literature: androstadienone.
Four findings at this tier reshape how the problem is understood.
Chemical structure does not predict brain activity; perception does. A distance matrix built from 32 chemical structural features does not align significantly with brain data in any region, whereas the perceptual distance matrix aligns with the posterior piriform cortex. The “one molecule, one brain effect” model is refuted right at the level of the primary olfactory cortex.
CO₂ — an odourless substance — activates the piriform cortex with every measurement method, from PET to coordinate-based meta-analysis to independent component analysis. The consequence: seeing the piriform light up does not prove that it is an olfactory response. Much of the “brain-imaging evidence” in essential-oil marketing material cannot rule out the possibility that the signal comes from the trigeminal nerve.
Trigeminal stimulation is not synonymous with unpleasantness. On a 0–100 scale, menthol was rated 75.4 for pleasantness with only 4.0 for pain; cinnamaldehyde at 75% by volume was rated 80.4 and 1.2. A substantial part of what is called the “relaxing effect of essential oils” may well be a pleasant cooling or warming sensation on the mucosa, not an odour effect. Notably, regular exposure increases trigeminal sensitivity rather than decreasing it.
Valence is not a property of the molecule. The same chemical labelled “cheddar cheese” and labelled “body odour” produces two different correlation patterns in the anterior cingulate cortex, mOFC and bilateral amygdala. Even at the olfactory bulb level, about 38.2% of the variance in the valence representation is individual variance. And an encoding model using a person’s own perceptual ratings predicts brain activity better than one using other people’s ratings. There is no “valence of molecule X”; there is only this person’s valence for molecule X in this context.
Two methodological limits apply to the whole tier. Sample sizes in the foundational studies run from single digits to the low double digits: 6 and 4 people in the studies decoding odour identity, 12 in three other studies, 14 and 15 in two more. And not one study reports the molecular concentration in inhaled air; every “concentration” is the concentration of the source solution, the mass in the vial, or a flow ratio. Doses cannot be converted from one study to another.
Scientific reference material, not medical advice and not promotional material. All assessments of evidence quality are the compilers’ own. Figures are redrawn from published data.