3.5 Orbitofrontal cortex: subjective value, labels and state
The OFC receives olfactory input directly from the piriform cortex and indirectly via the mediodorsal thalamus, making it a point of convergence between sensory signals and context. Sagar 2023, using dense data (≥4,320 trials per person, 160 odours), showed that the number of perceptual-space dimensions encoded increases progressively from piriform → amygdala → OFC, and that the OFC is the only region in which the “fine” code exceeds the “coarse” code in all three participants; an encoding model using the person’s own ratings predicted better than one using another person’s ratings (P = 0.016) — the OFC encodes odour experience in a personal way [12]. Sagar 2026 showed that the OFC represents valence most strongly (β = 0.569) and is the only region with significant negative cross-decoding between the two valence classes (r = −0.068), i.e. it has a shared “salience” axis [13].
Verbal labels. de Araujo 2005: the same test odour was rated −0.10 with the label “cheddar cheese” and −0.86 with the label “body odour” (t(11) = 6.68); mOFC/rACC tracked pleasantness, and notably, with clean air and only a label, the mOFC still responded whereas the amygdala did not [107]. Implication: the amygdala needs genuine sensory input, while the OFC integrates expectation.
Bodily state and reward learning. Selective satiation reduces OFC and amygdala responses to a CS+ predicting a devalued food odour, but not to one predicting a non-devalued odour [61]. Grabenhorst 2007 dissociated the medial OFC (pleasantness, r = 0.88) from the mid-posterior OFC/insula (unpleasantness) and the piriform cortex (intensity, r = 0.86), and showed that adding indole to jasmine made the medial OFC respond more strongly than to pure jasmine even though pleasantness ratings did not change (p > 0.3; participants were pre-selected) — a neural hint, not yet behavioural evidence, for the practice of using low doses of “unpleasant notes” in perfumery [53]. A right OFC lesion can abolish conscious awareness of odour while preserving sniff modulation and skin conductance [59].
Cognitive maps. Bao 2019 (n = 25) showed that when participants “navigate” a two-dimensional odour space, the vmPFC shows grid-cell-like 6-fold modulation and multivoxel patterns in the entorhinal cortex predict performance (r = 0.48) [54]. This is a conceptual bridge between the OFC/vmPFC (value) and the entorhinal cortex (relational space).
3.6 Entorhinal–hippocampus: association, memory and space
Privileged connectivity. Zhou 2021 (fMRI n = 25 + iEEG n = 8) showed that every subdivision of the olfactory cortex is more strongly connected with the hippocampus than are the visual, auditory and somatosensory cortices (F(3,72) = 25.8; P = 1.9×10−11), favouring the anterior hippocampus; piriform–hippocampal theta synchrony exceeded auditory–hippocampal synchrony in 8/8 patients and rose immediately after sniff onset [5]. This is the quantitative basis for the claim that odour is “closer to the hippocampus” than the other senses.
Temporal order. Kehl 2024 measured the onset of identity decoding: piriform 350 ms → amygdala 400 ms → entorhinal 850 ms → hippocampus 1,100 ms → parahippocampal 1,650 ms; decoding accuracy in the hippocampus correlated with the rate of correct odour identification (r = 0.50) [11]. Jiang 2017 (iEEG) showed that odour-evoked theta is locked to the sniff phase, with decoding onset at ~110 ms in PPC/amygdala and a peak at ~518 ms [51]. Yang 2021 found 15–30-fold compressed “replay” of odour sequences in the piriform cortex during a Sternberg task, more common before correct responses (75% vs 53%) [35].
First association. Yeshurun 2009: after one week, the first odour–object association was remembered better than the second (50.5% vs 34.5%), especially with unpleasant odours (56.3% vs 28.8%); the left hippocampus was more active when recalling the first association only for odours, not for sounds, and day-1 activity predicted day-8 memory (F = 14.6) [87]. Gottfried 2004: correctly recognising objects previously paired with an odour — even though the test contained no odour — reactivated the piriform cortex and anterior hippocampus [88]. Note: the behavioural advantage of the first association did not differ between odours and sounds (F = 0.03; p = 0.85); only the hippocampal signature was odour-specific [87]. Together the two results show that odours create multimodal memory traces with a distinctive hippocampal signature, suggesting a mechanism for the durability of childhood odour memories.
Circuits in animals (correlational and causal). Igarashi 2014 (tetrode recording, correlational): 20–40 Hz LEC–CA1 coherence increased progressively with odour–place associative learning and was absent on incorrect trials; the proportion of CA1 cells phase-locked to the LEC rose from 7.9% → 17.4% [90]. Lee 2021: inhibiting LEC layer IIa “fan cells” impaired learning of new odour pairs (success 90.9% → 25%) without affecting recall of old ones; 42% of fan cells carry D1 receptors, and inhibiting dopamine fibres in the LEC reproduced exactly the same pattern of deficit [91]. This is the most specific mechanism currently available for how value (dopamine) is written into the odour association map.
Spatial memory. Dahmani 2018: odour identification correlated with navigation performance (r = −0.37 with errors), mainly in people using spatial strategies (r = −0.57), via right hippocampal volume and left mOFC thickness; mOFC lesions reduced both [89]. However, the Szychowska 2025 meta-analysis (k = 7) showed that odour-based spatial memory is poorer than vision-based spatial memory (SMD −0.48; −0.77 to −0.19; I² = 71%) [36]. The two results do not conflict: olfaction and navigation share a substrate, but in humans odour is a poorer spatial cue than vision.
Structure and function. In healthy people (n = 117), only right hippocampal volume correlated — weakly — with odour threshold (r = 0.21; controlling for age r = 0.18) [38]. In patients with olfactory dysfunction, 90% of studies with a control group found reduced olfactory bulb volume; grey matter was reduced in the OFC, hippocampus, insula, amygdala and piriform cortex [37]. Bilateral amygdala damage (SM046) reduced odour recognition memory to chance (2AFC 10/20) while verbal memory was normal; temporal lobectomy, most clearly on the left, increased false alarms in odour–name matching (group effect η² = 0.23) [24].
3.7 Odour-evoked autobiographical memory: quantifying the “Proust effect”
The LOVER model (Limbic, Old, Vivid, Emotional, Rare) summarises the features thought to be specific to odour-evoked memories [86]. The small meta-analysis by Hackländer 2019 (13 studies, 14 experiments) allows each letter to be tested [32]:
| Attribute | Estimate | Test | Conclusion |
|---|---|---|---|
| Old — concentrated in the first decade of life | 51% vs 22% | F(2,8) = 16.2; ηp² = 0.80 | Strongly supported |
| Rare — rarely rehearsed | d = 0.52 | t(3) = 3.54; p = 0.019 | Supported (small k) |
| Emotional — emotional intensity | d = 0.47 | t(9) = 2.33; p = 0.022 | Supported |
| Feeling of being “brought back to the past” | d = 0.47 | t(6) = 2.36; p = 0.028 | Supported |
| More positive | d = 0.35 | t(9) = 3.43; p = 0.004 | Supported |
| Vivid — vividness | d = 0.10 | t(9) = 2.02; p = 0.037 | Significant but very small; not robust if one null study is assumed |
| Cueing efficiency (success rate) | 63% vs 76% | Z = 2.52; p = 0.012 | Contrary to LOVER: odours evoke fewer memories |
| Limbic | fMRI n = 5 (Herz 2004 [31]); Arshamian 2013 via [86] | Thin neural evidence | |
Two points deserve emphasis. (1) The emotional advantage is real but small to moderate (d ≈ 0.5), not the “large effect” often popularised; Willander & Larsson (n = 93 older adults) did not even find a clear emotional advantage, although they confirmed the “old” property [85]. (2) Odour is an inefficient but specific cue: once it does evoke a memory, the memory is richer. Chu & Downes 2002 showed that a congruent odour led participants to produce 120 additional new details (rate 0.60) compared with a label (0.47) and a picture (0.46), whereas an incongruent odour reduced output (0.36) — ruling out a “general arousal” explanation [84]. Adding the odour name increases retrieval but abolishes the childhood peak [86], consistent with the hypothesis that odour memories are encoded before language becomes dominant.
New evidence from animals provides a candidate mechanism: mice exposed to an odour during juvenile play retained a preference for it for up to 2 months, dependent on olfactory bulb granule cells born on day P1; without re-encountering the odour, the memory faded by 6 months [34]. A parallel survey in humans (n = 647) found that 73% of earliest odour memories came from events repeated >5 times [34] — suggesting that a “Proustian memory” is often a semantic–emotional memory of a repeated context rather than a single episode.
Scientific reference material, not medical advice. Figures are redrawn from published data, not copied from the original figures. GRADE and risk-of-bias ratings are the compilers’ own. Every number traces back to detailed reading notes on each paper.