Part II. Receptors: structure, biophysics and theories

2.1 The structural map 2018–2026

Volume 3 noted the 2023 structure of OR51E2 as the first atomic structure of a human odorant receptor. Seen as a whole, there is a pervasive limitation that has not been stated clearly: OR51E2 is a receptor for short-chain fatty acids, not a “typical” odorant receptor, and every other human odorant receptor structure is either a consensus receptor (an artificial sequence built from the average of the whole family) or has been back-mutated for stability.

2.2 AlphaFold and the limits of structure prediction

With about 400 receptors of which only a handful have experimental structures, structure-prediction models are an indispensable tool. AlphaFold2 predicts the core framework very well: a deviation of only 1.1 Å over 242 Cα atoms compared with the experimental structure of the consensus receptor OR52cs in the unbound state. But there are three limitations. Odorant receptors share less than 20% sequence identity with GPCRs of known structure; AlphaFold returns the unbound state with a closed pocket; and the flexible extramembrane regions have low confidence. The practical consequence: a 2026 conference report (only the abstract could be read, with no data) claims that AI docking models predict odorant receptor activation by novel compounds at close to chance level — in contrast with AI models predicting crowd perception, which do quite well.

2.3 Transduction biophysics: human data are almost absent

Most of what textbooks say about ion currents in olfactory neurons comes from frogs and mice. Direct data on human olfactory neurons consist only of semi-quantitative calcium imaging from biopsies: about 52% of biopsy samples yielded neurons responsive to odours, about 24% of neurons responded, and notably no neuron responded to more than one odorant Human. No patch-clamp recording has published single-channel current amplitudes in human olfactory neurons.

2.4 Shape or vibration?

Shape theory holds that receptors recognise molecules by their geometry and physicochemical properties. Vibration theory holds that receptors “read” the vibrational frequencies of chemical bonds. The key test is isotopic: replacing hydrogen with deuterium changes vibration while leaving shape almost unchanged.

Current consensus leans strongly towards shape and physicochemical theory. The 2023–2026 structures reinforce this: they show odorants held by specific chemical interactions in the binding pocket, with no need for a vibration-reading mechanism. The musk result of Gane 2013 has still not been fully explained, but it stands alone and contradicts the acetophenone result in the same paper.

2.5 Chirality: humans discriminate enantiomers less well than we think

The example of carvone — the R form smelling of spearmint, the S form of dill — is often used to illustrate that the human nose can tell the “left and right hands” of a molecule. The data show that carvone is the exception rather than the rule. Tested in 20 people with 10 enantiomer pairs, at group level only 3/10 pairs were discriminated above chance: α-pinene, carvone, limonene. The other seven pairs, including menthol, rose oxide and camphor, could not be discriminated Human. At the individual level, some people discriminated as many as 6/10 pairs.

A study of rose oxide showed a striking phenomenon: the two enantiomers produced different electrical responses at the epithelium (6.42 versus 3.77 µV; p = 0.04) but this difference did not reach the cortex, and psychophysically people could not discriminate them (3.2/9 correct trials, p = 0.41). Chiral information may be lost between the periphery and the centre.

A further note from Volume 3: for linalool, the two enantiomers have detection thresholds that differ three- to fourfold, so any comparison of “the effects of the two enantiomers” must first match intensity — a condition that the most-cited study on this topic does not meet.

2.6 The combinatorial code measured in human receptors

The largest screen of human receptors used 511 receptor lines, covering 394 of 418 intact receptor genes (94%). Of the 1,572 receptor–compound pairs tested, 425 gave responses, corresponding to 160 receptors. And 17 of the 27 receptors tested (63%) carried variants that alter function Human. The 2024 aggregated database M2OR gathers about 51 thousand receptor–compound pairs from 11 species, about two thirds of them human; the online version now exceeds 53 thousand.

The most important gap: there is no standard figure for combinatorial size at physiological concentrations — how many receptors an odorant activates at realistic airborne concentrations. In vitro screens use micromolar to millimolar concentrations, many orders of magnitude higher than in reality, and so may exaggerate combinatorial size.

2.7 Machine learning to predict odour

The 2023 Principal Odor Map, built with a graph neural network on about 5,000 molecules, achieved an AUROC of 0.89 and predicted as well as or better than the median human rater for 53% of molecules (random-forest model: 41%). Distances on the map correlated R = 0.73 with perceptual distances.

Two details are rarely mentioned but important. First, when the quality of 50 stimulus samples was checked, only 12 samples had an odour genuinely caused by the nominal compound itself; 18 were due to impurities, 16 to both, and 4 were unclear — a warning for every study using commercial chemicals. Second, the map predicts crowd descriptor labels well but does not predict differences between individuals.

A monograph compiled from 259 sourced facts collected under a common protocol. Practical-implication sections are the compilers’ inferences. Diagrams are newly drawn from published data. Scientific reference material, not medical advice.