Skip to main content
SpringerLink
Log in

High-precision mapping reveals the structure of odor coding in the human brain

  • Article
  • Published:

From Nature Neuroscience

View current issue Submit your manuscript

Abstract

Odor perception is inherently subjective. Previous work has shown that odorous molecules evoke distributed activity patterns in olfactory cortices, but how these patterns map on to subjective odor percepts remains unclear. In the present study, we collected neuroimaging responses to 160 odors from 3 individual subjects (18 h per subject) to probe the neural coding scheme underlying idiosyncratic odor perception. We found that activity in the orbitofrontal cortex (OFC) represents the fine-grained perceptual identity of odors over and above coarsely defined percepts, whereas this difference is less pronounced in the piriform cortex (PirC) and amygdala. Furthermore, the implementation of perceptual encoding models enabled us to predict olfactory functional magnetic resonance imaging responses to new odors, revealing that the dimensionality of the encoded perceptual spaces increases from the PirC to the OFC. Whereas encoding of lower-order dimensions generalizes across subjects, encoding of higher-order dimensions is idiosyncratic. These results provide new insights into cortical mechanisms of odor coding and suggest that subjective olfactory percepts reside in the OFC.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1: Neural activity patterns in olfactory brain areas represent odor stimuli.
Fig. 2: Neural activity patterns represent odor percepts.
Fig. 3: Neural activity patterns represent fine-grained odor percepts.
Fig. 4: Modeling odor-evoked activity using individual perceptual spaces.
Fig. 5: Encoding of idiosyncratic perceptual spaces in the OFC.

Similar content being viewed by others

Data availability

Data to reproduce the main findings presented in this manuscript (odor database, perceptual ratings, ROIs and odor-evoked responses) are available at https://github.com/viveksgr/NEMO_scripts. Raw data are available upon access request at https://doi.org/10.5281/zenodo.7636722 (ref. 76). The timeframe for response to requests is approximately 10 business days. Molecular information of odors is accessible from a publicly available dataset from previous studies17,60.

Code availability

Code for preprocessing and reproducing the results presented in this manuscript is available at https://github.com/viveksgr/NEMO_scripts.

References

  1. Leopold, D. A., Wilke, M., Maier, A. & Logothetis, N. K. Stable perception of visually ambiguous patterns. Nat. Neurosci. 5, 605–609 (2002).

    CAS  PubMed  Google Scholar 

  2. Walsh, V. & Kulikowski, J. Perceptual Constancy: Why Things Look As They Do (Cambridge Univ. Press, 1998).

  3. Hudson, R. From molecule to mind: the role of experience in shaping olfactory function. J. Comp. Physiol. A 185, 297–304 (1999).

    CAS  PubMed  Google Scholar 

  4. Sell, C. S. On the unpredictability of odor. Angew. Chem. Int. Ed. 45, 6254–6261 (2006).

    CAS  Google Scholar 

  5. Mainland, J. D. et al. The missense of smell: functional variability in the human odorant receptor repertoire. Nat. Neurosci. 17, 114–120 (2014).

    CAS  PubMed  Google Scholar 

  6. Bushdid, C., Magnasco, M. O., Vosshall, L. B. & Keller, A. Humans can discriminate more than 1 trillion olfactory stimuli. Science 343, 1370–1372 (2014).

    CAS  PubMed  PubMed Central  Google Scholar 

  7. Koulakov, A., Kolterman, B. E., Enikolopov, A. & Rinberg, D. In search of the structure of human olfactory space. Front. Syst. Neurosci. 5, 65 (2011).

    PubMed  PubMed Central  Google Scholar 

  8. Shepherd, G. M. & Greer, C. A. in The Synaptic Organization of the Brain (ed. Shepherd, G. M.) 159–203 (Oxford Univ. Press, 1998).

  9. Miura, K., Mainen, Z. F. & Uchida, N. Odor representations in olfactory cortex: distributed rate coding and decorrelated population activity. Neuron 74, 1087–1098 (2012).

    CAS  PubMed  PubMed Central  Google Scholar 

  10. Pashkovski, S. L. et al. Structure and flexibility in cortical representations of odour space. Nature 583, 253–258 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  11. Stettler, D. D. & Axel, R. Representations of odor in the piriform cortex. Neuron 63, 854–864 (2009).

    CAS  PubMed  Google Scholar 

  12. Sosulski, D. L., Bloom, M. L., Cutforth, T., Axel, R. & Datta, S. R. Distinct representations of olfactory information in different cortical centres. Nature 472, 213–216 (2011).

    CAS  PubMed  PubMed Central  Google Scholar 

  13. Miyamichi, K. et al. Cortical representations of olfactory input by trans-synaptic tracing. Nature 472, 191–196 (2011).

    CAS  PubMed  Google Scholar 

  14. Roland, B., Deneux, T., Franks, K. M., Bathellier, B. & Fleischmann, A. Odor identity coding by distributed ensembles of neurons in the mouse olfactory cortex. eLife 6, e26337 (2017).

    PubMed  PubMed Central  Google Scholar 

  15. Howard, J. D., Plailly, J., Grueschow, M., Haynes, J. D. & Gottfried, J. A. Odor quality coding and categorization in human posterior piriform cortex. Nat. Neurosci. 12, 932–938 (2009).

    CAS  PubMed  PubMed Central  Google Scholar 

  16. Bolding, K. A. & Franks, K. M. Complementary codes for odor identity and intensity in olfactory cortex. eLife 6, e22630 (2017).

    PubMed  PubMed Central  Google Scholar 

  17. Keller, A. et al. Predicting human olfactory perception from chemical features of odor molecules. Science 355, 820–826 (2017).

    CAS  PubMed  PubMed Central  Google Scholar 

  18. Castro, J. B., Ramanathan, A. & Chennubhotla, C. S. Categorical dimensions of human odor descriptor space revealed by non-negative matrix factorization. PLoS ONE 8, e73289 (2013).

    CAS  PubMed  PubMed Central  Google Scholar 

  19. Bizley, J. K. & Cohen, Y. E. The what, where and how of auditory-object perception. Nat. Rev. Neurosci. 14, 693–707 (2013).

    CAS  PubMed  PubMed Central  Google Scholar 

  20. Lescroart, M. D. & Gallant, J. L. Human scene-selective areas represent 3D configurations of surfaces. Neuron 101, 178–192. e177 (2019).

    CAS  PubMed  Google Scholar 

  21. Grunewald, A. & Skoumbourdis, E. K. The integration of multiple stimulus features by V1 neurons. J. Neurosci. 24, 9185–9194 (2004).

    CAS  PubMed  PubMed Central  Google Scholar 

  22. Cichy, R. M., Pantazis, D. & Oliva, A. Resolving human object recognition in space and time. Nat. Neurosci. 17, 455–462 (2014).

    CAS  PubMed  PubMed Central  Google Scholar 

  23. Kriegeskorte, N., Mur, M. & Bandettini, P. Representational similarity analysis—connecting the branches of systems neuroscience. Front. Syst. Neurosci. 2, 4 (2008).

    PubMed  PubMed Central  Google Scholar 

  24. Wu, M. C., David, S. V. & Gallant, J. L. Complete functional characterization of sensory neurons by system identification. Annu. Rev. Neurosci. 29, 477–505 (2006).

    CAS  PubMed  Google Scholar 

  25. Allen, E. J. et al. A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence. Nat. Neurosci. 25, 116–126 (2022).

    CAS  PubMed  Google Scholar 

  26. Naselaris, T., Prenger, R. J., Kay, K. N., Oliver, M. & Gallant, J. L. Bayesian reconstruction of natural images from human brain activity. Neuron 63, 902–915 (2009).

    CAS  PubMed  PubMed Central  Google Scholar 

  27. Kay, K. N., Naselaris, T., Prenger, R. J. & Gallant, J. L. Identifying natural images from human brain activity. Nature 452, 352–355 (2008).

    CAS  PubMed  PubMed Central  Google Scholar 

  28. Snitz, K. et al. Predicting odor perceptual similarity from odor structure. PLoS Comput. Biol. 9, e1003184 (2013).

    CAS  PubMed  PubMed Central  Google Scholar 

  29. Yeshurun, Y. & Sobel, N. An odor is not worth a thousand words: from multidimensional odors to unidimensional odor objects. Annu. Rev. Psychol. 61, 219–241 (2010).

    PubMed  Google Scholar 

  30. Sirotin, Y. B., Shusterman, R. & Rinberg, D. Neural coding of perceived odor intensity. eNeuro https://doi.org/10.1523/ENEURO.0083-15.2015 (2015).

  31. Winston, J. S., Gottfried, J. A., Kilner, J. M. & Dolan, R. J. Integrated neural representations of odor intensity and affective valence in human amygdala. J. Neurosci. 25, 8903–8907 (2005).

    CAS  PubMed  PubMed Central  Google Scholar 

  32. Anderson, A. K. et al. Dissociated neural representations of intensity and valence in human olfaction. Nat. Neurosci. 6, 196–202 (2003).

    CAS  PubMed  Google Scholar 

  33. Lapid, H. et al. Neural activity at the human olfactory epithelium reflects olfactory perception. Nat. Neurosci. 14, 1455–1461 (2011).

    CAS  PubMed  Google Scholar 

  34. Gratton, C., Nelson, S. M. & Gordon, E. M. Brain-behavior correlations: two paths toward reliability. Neuron 110, 1446–1449 (2022).

    CAS  PubMed  Google Scholar 

  35. Rosenberg, M. D. & Finn, E. S. How to establish robust brain–behavior relationships without thousands of individuals. Nat. Neurosci. 25, 835–837 (2022).

    CAS  PubMed  Google Scholar 

  36. Gottfried, J. A., Winston, J. S. & Dolan, R. J. Dissociable codes of odor quality and odorant structure in human piriform cortex. Neuron 49, 467–479 (2006).

    CAS  PubMed  Google Scholar 

  37. Fournel, A., Ferdenzi, C., Sezille, C., Rouby, C. & Bensafi, M. Multidimensional representation of odors in the human olfactory cortex. Hum. Brain Mapp. 37, 2161–2172 (2016).

    CAS  PubMed  PubMed Central  Google Scholar 

  38. Zelano, C., Mohanty, A. & Gottfried, J. A. Olfactory predictive codes and stimulus templates in piriform cortex. Neuron 72, 178–187 (2011).

    CAS  PubMed  PubMed Central  Google Scholar 

  39. Sobel, N. et al. Sniffing and smelling: separate subsystems in the human olfactory cortex. Nature 392, 282–286 (1998).

    CAS  PubMed  Google Scholar 

  40. Kareken, D. A. et al. Olfactory system activation from sniffing: effects in piriform and orbitofrontal cortex. NeuroImage 22, 456–465 (2004).

    PubMed  Google Scholar 

  41. Cohen, L., Rothschild, G. & Mizrahi, A. Multisensory integration of natural odors and sounds in the auditory cortex. Neuron 72, 357–369 (2011).

    CAS  PubMed  Google Scholar 

  42. Olofsson, J. K., Zhou, G., East, B. S., Zelano, C. & Wilson, D. A. Odor identification in rats: behavioral and electrophysiological evidence of learned olfactory-auditory associations. eNeuro https://doi.org/10.1523/ENEURO.0102-19.2019 (2019).

  43. Li, W., Howard, J. D., Parrish, T. B. & Gottfried, J. A. Aversive learning enhances perceptual and cortical discrimination of indiscriminable odor cues. Science 319, 1842–1845 (2008).

    CAS  PubMed  PubMed Central  Google Scholar 

  44. Bae, J., Yi, J.-Y. & Moon, C. Odor quality profile is partially influenced by verbal cues. PLoS ONE 14, e0226385 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

  45. Wang, P. Y. et al. Transient and persistent representations of odor value in prefrontal cortex. Neuron 108, 209–224. e206 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

  46. Kriegeskorte, N. & Douglas, P. K. Interpreting encoding and decoding models. Curr. Opin. Neurobiol. 55, 167–179 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

  47. Schulze, P., Bestgen, A.-K., Lech, R. K., Kuchinke, L. & Suchan, B. Preprocessing of emotional visual information in the human piriform cortex. Sci. Rep. 7, 9191 (2017).

    PubMed  PubMed Central  Google Scholar 

  48. Wilson, C. D., Serrano, G. O., Koulakov, A. A. & Rinberg, D. A primacy code for odor identity. Nat. Commun. 8, 1477 (2017).

    PubMed  PubMed Central  Google Scholar 

  49. Gottfried, J. A., O’Doherty, J. & Dolan, R. J. Encoding predictive reward value in human amygdala and orbitofrontal cortex. Science 301, 1104–1107 (2003).

    CAS  PubMed  Google Scholar 

  50. Bowman, N. E., Kording, K. P. & Gottfried, J. A. Temporal integration of olfactory perceptual evidence in human orbitofrontal cortex. Neuron 75, 916–927 (2012).

    CAS  PubMed  PubMed Central  Google Scholar 

  51. Zhou, J. et al. Rat orbitofrontal ensemble activity contains multiplexed but dissociable representations of value and task structure in an odor sequence task. Curr. Biol. 29, 897–907. e893 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

  52. Bao, X., Raguet, L. L., Cole, S. M., Howard, J. D. & Gottfried, J. A. The role of piriform associative connections in odor categorization. eLife 5, e13732 (2016).

    PubMed  PubMed Central  Google Scholar 

  53. Yoshida, I. & Mori, K. Odorant category profile selectivity of olfactory cortex neurons. J. Neurosci. 27, 9105–9114 (2007).

    CAS  PubMed  PubMed Central  Google Scholar 

  54. De Araujo, I. E., Rolls, E. T., Velazco, M. I., Margot, C. & Cayeux, I. Cognitive modulation of olfactory processing. Neuron 46, 671–679 (2005).

    PubMed  Google Scholar 

  55. Keller, A. Attention and olfactory consciousness. Front. Psychol. 2, 380 (2011).

    PubMed  PubMed Central  Google Scholar 

  56. Li, W. et al. Right orbitofrontal cortex mediates conscious olfactory perception. Psychol. Sci. 21, 1454–1463 (2010).

    PubMed  Google Scholar 

  57. Tanabe, T., Iino, M. & Takagi, S. Discrimination of odors in olfactory bulb, pyriform-amygdaloid areas, and orbitofrontal cortex of the monkey. J. Neurophysiol. 38, 1284–1296 (1975).

    CAS  PubMed  Google Scholar 

  58. Potter, H. & Butters, N. An assessment of olfactory deficits in patients with damage to prefrontal cortex. Neuropsychologia 18, 621–628 (1980).

    CAS  PubMed  Google Scholar 

  59. Jones-Gotman, M. & Zatorre, R. J. Olfactory identification deficits in patients with focal cerebral excision. Neuropsychologia 26, 387–400 (1988).

    CAS  PubMed  Google Scholar 

  60. Keller, A. & Vosshall, L. B. Olfactory perception of chemically diverse molecules. BMC Neurosci. 17, 55 (2016).

    PubMed  PubMed Central  Google Scholar 

  61. Cai, M. B., Schuck, N. W., Pillow, J. W. & Niv, Y. Representational structure or task structure? Bias in neural representational similarity analysis and a Bayesian method for reducing bias. PLoS Comput. Biol. 15, e1006299 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

  62. Zhou, G., Lane, G., Cooper, S. L., Kahnt, T. & Zelano, C. Characterizing functional pathways of the human olfactory system. eLife 8, e47177 (2019).

    CAS  PubMed  PubMed Central  Google Scholar 

  63. Echevarria-Cooper, S. L. et al. Mapping the microstructure and striae of the human olfactory tract with diffusion MRI. J. Neurosci. 42, 58–68 (2021).

  64. Zald, D. H. & Pardo, J. V. Emotion, olfaction, and the human amygdala: amygdala activation during aversive olfactory stimulation. Proc. Natl Acad. Sci. USA 94, 4119–4124 (1997).

    CAS  PubMed  PubMed Central  Google Scholar 

  65. Zatorre, R. J., Jones-Gotman, M., Evans, A. C. & Meyer, E. Functional localization and lateralization of human olfactory cortex. Nature 360, 339–340 (1992).

    CAS  PubMed  Google Scholar 

  66. Burton, S., Murphy, D., Qureshi, U., Sutton, P. & O’Keefe, J. Combined lesions of hippocampus and subiculum do not produce deficits in a nonspatial social olfactory memory task. J. Neurosci. 20, 5468–5475 (2000).

    CAS  PubMed  PubMed Central  Google Scholar 

  67. Welvaert, M. & Rosseel, Y. On the definition of signal-to-noise ratio and contrast-to-noise ratio for fMRI data. PLoS ONE 8, e77089 (2013).

    CAS  PubMed  PubMed Central  Google Scholar 

  68. Tzourio-Mazoyer, N. et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 15, 273–289 (2002).

    CAS  PubMed  Google Scholar 

  69. Howard, J. D. & Kahnt, T. Identity prediction errors in the human midbrain update reward-identity expectations in the orbitofrontal cortex. Nat. Commun. 9, 1611 (2018).

    PubMed  PubMed Central  Google Scholar 

  70. Suarez, J. A., Howard, J. D., Schoenbaum, G. & Kahnt, T. Sensory prediction errors in the human midbrain signal identity violations independent of perceptual distance. eLife 8, e43962 (2019).

    PubMed  PubMed Central  Google Scholar 

  71. Shanahan, L. K., Bhutani, S. & Kahnt, T. Olfactory perceptual decision-making is biased by motivational state. PLoS Biol. 19, e3001374 (2021).

    CAS  PubMed  PubMed Central  Google Scholar 

  72. Noto, T., Zhou, G., Schuele, S., Templer, J. & Zelano, C. Automated analysis of breathing waveforms using BreathMetrics: a respiratory signal processing toolbox. Chem. Senses 43, 583–597 (2018).

    PubMed  PubMed Central  Google Scholar 

  73. Nili, H. et al. A toolbox for representational similarity analysis. PLoS Comput. Biol. 10, e1003553 (2014).

    PubMed  PubMed Central  Google Scholar 

  74. Esfahlani, F. Z. et al. High-amplitude cofluctuations in cortical activity drive functional connectivity. Proc. Natl Acad. Sci. USA 117, 28393–28401 (2020).

    CAS  Google Scholar 

  75. Gusfield, D. & Irving, R. W. The Stable Marriage Problem: Structure and Algorithms (MIT Press, 1989).

  76. Sagar, V., Shanahan, L. K., Zelano, C. M., Gottfried, J. A. & Kahnt, T. Neural encoding models of olfaction (NEMO) dataset. Zenodo https://doi.org/10.5281/zenodo.7636722 (2023).

Download references

Acknowledgements

We thank T. Parrish for help with optimizing the scanning sequence; H. Siddiqui, S. Attanti, D. Smith and R. Reynolds for help with data acquisition; and the Rockefeller University and Sage Bionetworks-DREAM for the database on molecular properties of odors. This work was supported by grants from the National Institute of Mental Health (grant no. T32 MH067564 to V.S.), the National Institute of Neurological Disorders and Stroke (grant no. T32 NS047987 to V.S.) and the Intramural Research Program at the National Institute on Drug Abuse (grant no. ZIA DA000642). The opinions expressed in this work are the authors’ own and do not reflect the view of the National Institutes of Health/Department of Health and Human Services.

Author information

Authors and Affiliations

  1. Department of Neurology, Feinberg School of Medicine, Northwestern University, Chicago, IL, USA

    Vivek Sagar & Christina M. Zelano

  2. Department of Psychology, Rhodes College, Memphis, TN, USA

    Laura K. Shanahan

  3. Department of Neurology, University of Pennsylvania, Philadelphia, PA, USA

    Jay A. Gottfried

  4. Department of Psychology, University of Pennsylvania, Philadelphia, PA, USA

    Jay A. Gottfried

  5. National Institute on Drug Abuse Intramural Research Program, Baltimore, MD, USA

    Thorsten Kahnt

Authors
  1. Vivek Sagar
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Laura K. Shanahan
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Christina M. Zelano
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Jay A. Gottfried
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Thorsten Kahnt
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

V.S. and T.K. conceived and designed the experiment. V.S. created the odorant stimuli and performed the experiments. V.S. and T.K. conceptualized the computational analyses. V.S. performed the data analysis. All the authors discussed the results and wrote and edited the manuscript.

Corresponding author

Correspondence to Thorsten Kahnt.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature Neuroscience thanks Mingbo Cai, Tali Weiss and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Perceptual odor descriptors and ratings.

a, Reliability of perceptual ratings. In each subject and for each descriptor, reliability of the perceptual descriptor is computed by correlating perceptual ratings for the same odor acquired in different sessions. Gray line indicates threshold for statistical significance (r > 0.131, threshold p = 0.05, n = 3 subjects, 160 odors/subject, one tailed t-test) and dots are individual subjects. Reliability is computed between different fMRI sessions for S1. For S2 and S3, the average ratings acquired in two behavioral sessions outside the scanner were correlated with ratings acquired inside the scanner (S2, r = 0.589; S3, r = 0.660, n = 3 subjects, 160 odors/subject). The correlation of odor-wise descriptor ratings (averaged across odors) between S2 and S3 was 0.377. b, Histogram of discriminability of odors for the average subject. Discriminability between two odors is the absolute difference (in standard deviations) of the perceptual feature with maximum difference. c, Perceptual similarity matrices for all subjects. Each cell in the matrix depicts the correlation between the perceptual ratings of two odors. For illustration, rows and columns are sorted using k-means, independently for each subject. d, Generalizability of perceptual ratings across subjects is computed as the correlation between the (off-diagonal entries of) the perceptual similarity matrices of two subjects and averaged across all subject pairs (r = 0.168, p = 0.0000, n = 3 subjects, 12720 odor pairs/subject, two-tailed t-test). Dots indicate subject pairs. The gray line indicates the threshold for statistical significance (r > 0.022, threshold p = 0.05, n = 3 subjects, 12720 odor pairs/subject, two-tailed t-test). Error bars indicate 95% C.I.

Extended Data Fig. 2 Neural responses to odors.

a, Task design comprising of self-paced behavioral task (top-panel) to acquire at least two sets of ratings per odor per descriptor and fMRI task (bottom panel) to rate the odors. S1 provided ratings in all fMRI sessions, whereas S2 and S3 did not rate odors in the third fMRI session. b, Odor-evoked fMRI response in each ROI for each subject. Shaded areas depict 95% C.I. for the mean (black lines) per subject. Peaks in all areas occurred at least 4 seconds after odor presentation. Analyses were restricted to up to 6 seconds to avoid confounding the neural activity with the perceptual rating task. For OFC in S3, BOLD response does not return to baseline, highlighting individual and inter-regional variability in the shape of the hemodynamic response. c, Mean percentage of gray matter voxels with significant odor-evoked responses for each ROI. Error bars depict 95% C.I. and lines depict individual subjects (n = 3 subjects, 160 odors/subject). d, Average temporal signal to noise ratio (t-snr: mean/standard deviation of the voxel time-series) in an ROI. Bars denote mean effects and errorbars are s.e.m. across subjects (n = 3 subjects, 160 odors/subject). t-snr did not differ significantly across areas (F3,8 = 0.39, p = 0.78, one way ANOVA). e, Neural similarity matrices for each ROI in each subject. Each cell in the matrix depicts the correlation between the multi-voxel response patterns of two odors. For illustration purposes, rows and columns are sorted using k-means (4 total clusters), independently for each subject. f, Correlation of neural activity patterns evoked by the same odor in different sessions (pattern reliability), averaged across odors and subjects. Error bars indicate s.e.m. across subjects. Pattern reliability is significant in all areas and in all subjects (r > 0, p = 0.000, Wilcoxon signed rank test, n = 3 subjects, 12720 odor pairs/subject), except PirF in S3 (r = 0.04, p = 0.086, Wilcoxon signed rank test, n = 3 subjects, 12720 odor pairs/subject). g, Pattern reliability separately measured between sessions 1 and 2, sessions 2 and 3, and sessions 1 and 3. Pattern reliability between sessions 1 and 2 and 2 and 3 is not significantly different from pattern reliability between sessions 1 and 3 (F1,6 = 0.02, p = 0.90, repeated measures 2-way ANOVA with session pairs and ROI as factors). There was no significant main effect of ROI (F3,6 = 2.07, p = 0.206), and no significant interaction (F3,6 = 2.12, p = 0.198), suggesting that odor-evoked activity patterns remained stable across fMRI sessions. Error bars indicate 95% C.I. For all tests, n = 3 subjects, 12720 odor pairs/subject.

Extended Data Fig. 3 Representational similarity analysis (RSA) for individual subjects.

RSA analysis based on coarse and fine-grained perceptual similarity for individual subjects. Correlations were taken across 12,720 odor pairs. a, Bars depict the Spearman rank-correlation between neural and coarse perceptual similarity (rc hatched) or fine-grained perceptual similarity matrices (rf solid), for individual subjects. Bars indicate mean correlation and error bars depict 95% C.I. (perc. bootstrap). In all subjects, fine-grained and coarse perceptual representational similarity is significant in AMY and OFC. In subject S1, representation of fine-grained perceptual similarity is significantly higher than coarse perceptual similarity in OFC, but not in any other area (PirF, rc = 0.010, p = 0.132, rf = 0.012, p = 0.075 p(rf>rc)=0.692; PirT, rc = 0.016, p = 0.022, rf = 0.017, p = 0.019, p(rf>rc)=0.932; AMY, rc = 0.018, p = 0.002, rf = 0.025, p = 0.0000, p(rf > rc) = 0.127; OFC, rc = 0.037, p = 0.0000, rf = 0.061, p = 0.0000, p(rf > rc) = 0.0000; A1, rc = 0.005, p = 0.472, rf = −0.0001, p = 0.988, p(rf > rc) = 0.312; wm, rc = 0.006, p = 0.351, rf = −0.002, p = 0.721, p(rf > rc) = 0.057, two-tailed bootstrap comparison). In subject S2, representation of fine-grained perceptual similarity is significantly higher than coarse perceptual similarity in OFC, but not in other areas (PirF, rc = -0.005, p = 0.442, rf = -0.013, p = 0.050, p(rf > rc) = 0.060; PirT, rc = 0.002, p = 0.793, rf = 0.009, p = 0.223, p(rf > rc) = 0.095; AMY, rc = 0.026, p = 0.0000, rf = 0.030, p = 0.0000, p(rf > rc) = 0.290; OFC, rc = 0.051, p = 0.0000, rf = 0.067, p = 0.0000, p(rf > rc) = 0.0000; A1, rc = 0.012, p = 0.076, rf = 0.016, p = 0.015, p(rf > rc) = 0.290; wm, rc = -0.003, p = 0.619, rf = -0.006, p = 0.330, p(rf > rc) = 0.463, two-tailed bootstrap comparison). In subject S3, representation of fine-grained perceptual similarity is significantly higher than coarse perceptual similarity in PirT, AMY and OFC, but not in PirF, A1 and wm (PirF, rc = 0.007, p = 0.295, rf = 0.0002, p = 0.960, p(rf > rc) = 0.177; PirT, rc = 0.026, p = 0.0000, rf = 0.039, p = 0.0000, p(rf > rc) = 0.009; AMY, rc = 0.030, p = 0.0000, rf = 0.041, p = 0.0000, p(rf > rc) = 0.018; OFC, rc = 0.101, p = 0.0000, rf = 0.122, p = 0.0000, p(rf > rc) = 0.0000; A1, rc = −0.0002, p = 0.976, rf = 0.005, p = 0.460, p(rf > rc) = 0.282; wm, rc = 0.005, p = 0.476, rf = −0.002, p = 0.771, p(rf > rc) = 0.553, two-tailed bootstrap comparison). Thus, OFC is the only ROI where the fine-grained RSA exceeds the coarse RSA in all three subjects. b, Difference between the neural representation of fine-grained and coarse perceptual similarity in a (r). Bars depict mean correlation difference in each subject, error bars depict 95% C.I. (perc. bootstrap). The difference is significantly larger in OFC than in PirF in all subjects (OFC-PirF all subjects, p = 0.0000), in PirT for S1 (p = 0.0012) but not S2 (p = 0.106) or S3 (p = 0.211) and in AMY for S1 (p = 0.0012) and S2 (p = 0.025) but not in S3 (p = 0.171) (two-tailed bootstrap comparison, 12720 odor pairs). The difference between the coarse and fine-grained RSA is maximum in OFC across areas for all subjects. Further, OFC is the only area where the difference between the coarse and fine-grained RSA is significant across all subjects.

Extended Data Fig. 4 Control analyses for RSA.

We performed control RSAs in olfactory ROIs as well as control areas A1 (primary auditory cortex) and wm (white matter voxels). For statistics on subject-wise results, see Supplementary Table 2. a, (Top panel) bars depict the Spearman rank-correlation between neural and coarse (rc hatched) or fine-grained perceptual similarity matrices (rf solid), averaged across subjects, adjusted to include intensity and pleasantness. rf>rc in all areas except PirF, A1 and wm. All p-values are based on null hypothesis rc = rf, tested using two tailed bootstrap comparison (PirF, rc = 0.005, rf = 0.005, p = 0.992; PirT, rc = 0.022, rf = 0.035, p = 0.0000; AMY, rc = 0.040, rf = 0.059, p = 0.0000; OFC, rc = 0.084, rf = 0.120, p = 0.0000; A1, rc = 0.014, rf = 0.015, p = 0.734; wm, rc = 0.008, rf = 0.002, p = 0.03). Note that in wm, rc significantly exceeds rf (that is, rc > rf), which is the opposite of what is expected and found in olfactory brain areas, and testing rf > rc using a one-tailed test is not significant (p = 0.97). (Bottom panel) Difference between the fine-grained and coarse representational similarity in a, top panel (r). Difference is significantly higher in OFC than in PirF, PirT, AMY, A1 or wm (all areas, p = 0.0000, two-tailed bootstrap comparison). b, (Top panel) bars depict the Spearman rank correlation between neural and coarse (rc hatched) or fine-grained perceptual similarity matrices (rf solid), averaged across subjects, adjusted to account for differences in size of the ROI. 70 voxels were chosen with replacement from each ROI and subject to construct the neural similarity matrix. rf>rc only in the OFC and not other areas (PirF, rc = 0.004, rf = 0.000, p = 0.196; PirT, rc = 0.012, rf = 0.018, p = 0.080; AMY, rc = 0.019, rf = 0.026, p = 0.077; OFC, rc = 0.049, rf = 0.064, p = 0.0000; A1, rc = 0.005, rf = 0.006, p = 0.745; wm, rc = 0.002, rf = −0.002, p = 0.172). (Bottom panel) Difference between the fine-grained and coarse representational similarity in b, top panel (r). Difference is significantly higher in OFC than in PirF, A1 or wm (p = 0.0000) and trending for PirT (p = 0.053) and AMY (p = 0.074). c, (Top panel) bars depict the Spearman rank correlation between neural and coarse (rc hatched) or fine-grained perceptual similarity matrices (rf solid), averaged across subjects, adjusted to account for perceptual correlations with molecular features. 4869 molecular features were used to construct the molecular similarity matrix. Molecular similarity was regressed out from both fine-grained and coarse perceptual similarity matrices. rf>rc in all areas except PirF, A1 and wm (PirF, rc = 0.003, rf = -0.000, p = 0.113; PirT, rc = 0.013, rf = 0.020, p = 0.010; AMY, rc = 0.021, rf = 0.029, p = 0.002; OFC, rc = 0.058, rf = 0.078, p = 0.0000; A1, rc = 0.004, rf = 0.006, p = 0.538; wm, rc = 0.002, rf =−0.003, p = 0.060). (Bottom panel) Difference between the fine-grained and coarse representational similarity in c, top panel (r). Difference is significantly higher in OFC than all areas (PirF, A1, wm, p = 0.0000; PirT, p = 0.0004; AMY, p = 0.0002). d, (Top panel) bars depict the Spearman rank correlation between neural and coarse (rc hatched) or fine-grained perceptual similarity matrices (rf solid), averaged across subjects, after excluding odors with low detectability. rf>rc all areas except wm (PirF, rc = 0.002, rf = -0.007, p = 0.005; PirT, rc = 0.006, rf = 0.015, p = 0.008; AMY, rc = 0.021, rf = 0.030, p = 0.007; OFC, rc = 0.051, rf = 0.078, p = 0.0000; A1, rc = 0.000, rf = 0.087, p = 0.049; wm, rc = −0.001, rf = 0.000, p = 0.701). (Bottom panel) Difference between the fine-grained and coarse representational similarity in d, top panel (r). Difference is significantly higher in OFC than all areas (PirF, AMY, A1, wm p = 0.0000; PirT, p = 0.0001;). e, (Top panel) bars depict the Spearman rank correlation between neural and coarse (rc hatched) or fine-grained perceptual similarity matrices (rf solid), averaged across subjects when neural responses were extracted from the same time bin (5 second after odor onset) in all areas and subjects. rf > rc in PirT, AMY and OFC but not other areas (PirF, rc = -0.001, rf = 0.002, p = 0.184; PirT, rc = 0.013, rf = 0.020, p = 0.012; AMY, rc = 0.030, rf = 0.038, p = 0.002; OFC, rc = 0.060, rf = 0.081, p = 0.0000; A1, rc = 0.002, rf = 0.002, p = 0.859; wm, rc = 0.000, rf = 0.001, p = 0.733). (Bottom panel) Difference between the fine-grained and coarse representational similarity in e, top panel (r). Difference is significantly higher in OFC than all areas (all areas, p = 0.0000). f, (Top panel) bars depict the Pearson’s (instead of Spearman) correlation between neural and coarse (rc hatched) or fine-grained perceptual similarity matrices (rf solid), averaged across subjects. rf>rc only in PirT, AMY and OFC but not other areas (PirF, rc = 0.002, rf = 0.002, p = 0.97; PirT, rc = 0.016, rf = 0.022, p = 0.046; AMY, rc = 0.027, rf = 0.037, p = 0.002; OFC, rc = 0.070, rf = 0.093, p = 0.0000; A1, rc = 0.005, rf = 0.008, p = 0.360; wm, rc = 0.002, rf = 0.000, p = 0.364). (Bottom panel) Difference between the fine-grained and coarse representational similarity in f, top panel (r). Difference is significantly higher in OFC than all areas (all areas, p = 0.0000). For all panels, error bars depict 95% C.I. (perc. bootstrap) and comparisons are based on two tailed bootstrap comparison, n = 3 subjects, 12720 odor pairs/subject.

Extended Data Fig. 5 Statistical control analyses for RSA.

a, To account for potential statistical biases in the bootstrap procedure, we performed additional permutation tests for perceptual and molecular RSA effects (Fig. 2b). For this, we generated null distributions by randomly shuffling perceptual and molecular ratings across odors. Plots show the means and 95% C.I. for the null distributions of perceptual and molecular RSA effects, which were (as expected) not significantly different from zero in any area for any subject (p > 0.2, all areas, all subjects). Solid lines indicate 95% C.I. for perceptual RSA and dashed lines indicate 95% C.I. (two tailed percentile bootstrap) for molecular RSA. Importantly, we used these null distributions to compute p-values for the perceptual and molecular RSA shown in Fig. 2b, confirming that rp is significant in PirT (p = 0.0000), AMY (p = 0.0000), OFC (p = 0.0000), A1 (p = 0.008) but not PirF (p = 0.308) or wm (p = 0.733). Moreover, rp significantly exceeds rm in OFC (p = 0.0000) but not in PirF (p = 0.288), PirT (p = 0.102), AMY (p = 0.173), A1 (p = 0.99) or wm (0.741, two tailed permutation test). To further test for biases in the bootstrap approach, we tested whether the number of odor pairs selected in each bootstrap affects the results. That is, we computed the correlation between the number of unique odor pairs in each bootstrap and rp and rm which was not significant in most areas and subjects (all areas, p > 0.05, one sample t-test) except AMY in S1 (p = 0.035, one sample t-test). b, To account for potential statistical biases in the bootstrap procedure, we performed additional permutation tests for coarse and fine-grained perceptual RSA effects (Fig. 3b). Similar to the analysis described in panel a, we generated null distributions by randomly shuffling perceptual ratings across odors. Plots show the means and 95% C.I. (two tailed percentile bootstrap) for the null distributions of coarse and fine-grained perceptual RSA effects, which were (as expected) not significantly different from zero in any area for any subject (p > 0.2, all areas, all subjects). Solid lines indicate 95% C.I. for fine-grained RSA and dashed lines indicate 95% C.I. for coarse RSA. Importantly, we used these null distributions to compute p-values for the coarse and fine-grained perceptual RSA effects shown in Fig. 3b, confirming that rc is significant in PirT (p = 0.006), AMY (p = 0.0000), OFC (p = 0.0000), but not PirF (p = 0.401), A1 (p = 0.280) or wm (p = 0.589), whereas rf is significant PirT (p = 0.0003), AMY (p = 0.0000), OFC (p = 0.0000), but not PirF (p = 0.98), A1 (p = 0.182) or wm (p = 0.660). Moreover, rf > rc is significant in AMY (p = 0.0232), OFC (p = 0.0000) and trending in PirT (p = 0.051), but not significant in PirF (p = 0.198), A1 (p = 0.651) or wm (0.147, two tailed permutation test). c, To further validate our RSA results, we compared rc and rf in olfactory areas to rc and rf in our control area A1. All olfactory areas (except PirF) had significantly larger representational similarities for fine-grained (rf) odor percepts than A1 (difference between representational similarities in the ROI and A1 denoted by ROI-A1) (rc: PirF-A1, p = 0.794; PirT-A1, p = 0.058; AMY-A1, p = 0.0000; OFC-A1, p = 0.0000; wm-A1, p = 0.601; rf: PirF-A1, p = 0.161; PirT-A1, p = 0.002; AMY-A1, p = 0.0000; OFC-A1, p = 0.0000; wm-A1, p = 0.086, two tailed bootstrap comparison). For all panels, bars indicate mean effects and error bars depict 95% C.I. (perc. bootstrap), n = 3 subjects, 12720 odor pairs/subject.

Extended Data Fig. 6 RSA control analyses for intensity, pleasantness and sniff evoked activity.

a, We examined representational similarities based exclusively on intensity or pleasantness. The intensity RSA is significant in all areas (PirF, PirT, AMY, OFC, A1, p = 0.0000; wm, p = 0.033), while the pleasantness RSA is significant only in the olfactory areas: PirF, PirT, AMY and OFC but not A1 or wm (PirF, p = 0.002; PirT, AMY, OFC, p = 0.0000; A1, p = 0.105; wm, p = 0.42, two tailed bootstrap comparison, n = 3 subjects, 12720 odor pairs/subject). b, RSA results when intensity or pleasantness is regressed out of the perceptual descriptor ratings. Two RSA models were constructed: one without intensity and one without pleasantness. The RSA without intensity is significant in PirT, AMY and OFC but not PirF, A1 or wm (PirF, p = 0.345; PirT, p = 0.006; AMY, p = 0.0000; OFC, p = 0.0000; A1, p = 0.903; wm, p = 0.125). The RSA without pleasantness is significant in PirF, PirT, AMY, OFC, A1 but not wm (PirF, p = 0.039; PirT, AMY, OFC, A1, p = 0.0000; wm, p = 0.778, two tailed bootstrap comparison). This suggests that perceptual encoding does not exclusively rely on intensity and/or pleasantness in olfactory areas (PirT, AMY or OFC) and that RSA results in the A1 control area are exclusively driven by odor intensity. For all tests, n = 3 subjects, 12720 odor pairs/subject. c, Pearson’s correlation of intensity ratings and sniff volumes (averaged across all trials) across 160 odors for each subject. d, Pearson’s correlation of intensity ratings and sniff durations (averaged across all trials) across 160 odors for each subject. e, Regressing odor similarity based on sniff volume from intensity and pleasantness similarity and computing the residual RSA for intensity and pleasantness (similar to a). The intensity RSA is significant in all areas (PirF, PirT, AMY, OFC, A1, p = 0.0000, two-tailed boostrap comparison) except wm, p = 0.128, while the pleasantness RSA is significant only in the olfactory areas: PirF, PirT, AMY and OFC but not A1 or wm (PirF, p = 0.001; PirT, p = 0.002; AMY, OFC, p = 0.0000; A1, p = 0.639; wm, p = 0.543, n = 3 subjects, 12720 odor pairs/subject). f, Regressing odor similarity based on sniff duration from intensity and pleasantness similarity and computing the residual RSA for intensity and pleasantness (similar to a). The intensity RSA is significant in all areas (PirF, PirT, AMY, OFC, p = 0.0000; A1,p = 0.001, two tailed boostrap comparison) except wm, p = 0.392, while the pleasantness RSA is significant only in the olfactory areas: PirF, PirT, AMY and OFC but not A1 or wm (PirF, p = 0.005; PirT, p = 0.044; AMY, p = 0.004; OFC, p = 0.0000; A1, p = 0.616; wm, p = 0.792, n = 3 subjects, 12720 odor pairs/subject). g, We regressed odor similarity based on sniff volume from coarse and fine-grained perceptual similarity and computed the residual RSA. Results are similar to Fig. 3b. rf > rc in all areas except PirF, A1 and wm (PirF, rc = 0.002, rf = −0.001, p = 0.150; PirT, rc = 0.012, rf = 0.018, p = 0.019; AMY, rc = 0.021, rf = 0.029, p = 0.004; OFC, rc = 0.059, rf = 0.098, p = 0.0000; A1, rc = 0.003, rf = 0.004, p = 0.638; wm, rc = 0.002, rf = −0.003, p = 0.055). h, We regressed odor similarity based on sniff duration from coarse and fine-grained perceptual similarity and computed the residual RSA. Results are similar to Fig. 3b. rf > rc in all areas except PirF, A1 and wm (PirF, rc = 0.003, rf = -0.001, p = 0.090; PirT, rc = 0.012, rf = 0.018, p = 0.006; AMY, rc = 0.020, rf = 0.027, p = 0.003; OFC, rc = 0.057, rf = 0.077, p = 0.0000; A1, rc = 0.003, rf = 0.004, p = 0.680; wm, rc = 0.002, rf = −0.003, p = 0.063). In all panels, error bars indicate 95% C.I.

Extended Data Fig. 7 RSA for increasing numbers of perceptual descriptors.

A, Perceptual representational similarity as a function of the number of perceptual descriptors used in estimating perceptual similarity. The case when only 1 descriptor is used corresponds to coarse representational similarity while the case when 16 descriptors are used corresponds to fine-grained representational similarity (Fig. 3b). b, Slope of perceptual representational similarity as a function of number of perceptual descriptors used. Error bars are s.e.m. across subjects. Slopes are maximal for OFC in all subjects (F3,8 = 6.99, p = 0.013, one way ANOVA, n = 3 subjects). This indicates that fine-grained representational similarity in the OFC increases as additional descriptors are added in the model.

Extended Data Fig. 8 Control analyses for encoding models.

A, Mean prediction accuracy of the encoding model using 14 orthogonal principal components (explaining at least 90% of the variance) of the perceptual descriptors as basis functions. B, Percentage of odor-responsive gray matter voxels with significant prediction accuracy (threshold p = 0.05, one-tailed one-sample t-test, FDR corrected) with PCA basis. c, Dimensionality of encoding for the encoding model with PCA basis. Dimensionality of encoding increases from PirF to OFC (p = 0.000, FWE against the null hypothesis κ(PirF) = κ(PirT) = κ(AMY) = κ(OFC), two-tailed bootstrap comparison). d, Mean prediction accuracy of the encoding model with 4-fold cross-validation where training and test odors came from independent scanning sessions. e, Percentage of odor-responsive gray matter voxels with significant prediction accuracy (threshold p = 0.05, one-tailed one-sample t-test, FDR corrected) for encoding model with 4-fold cross-validation. f, Dimensionality of encoding for the encoding model with 4-fold cross-validation. Dimensionality of encoding increases from PirF to OFC (p = 0.000, FWE against the null hypothesis κ(PirF) = κ(PirT) = κ(AMY) = κ(OFC), two-tailed bootstrap comparison). g, Prediction accuracy of encoding model with shuffled perceptual ratings is not significant for any area in any subject (p > 0.1, all areas, all subjects, two tailed shuffle test). h, Mean prediction accuracy of the encoding model without odors with low detectability is significantly greater than zero in all ROIs and subjects (except PirF in subject S1, p = 0.65, PirF S3, p = 0.03, remaining areas/subjects p = 0.0000, two sided Wilcoxon signed rank test). These results are qualitatively similar to those obtained when odors with low detectability are included (Fig. 4c). i, Mean prediction accuracy of encoding model in primary auditory cortex (A1) and white matter (wm) (A1, mean r = 0.027; wm mean r = 0.045) are much lower than those observed in olfactory areas (Fig. 4c). j, Percentage of voxels in A1 and wm that show significant prediction accuracy (threshold p = 0.05, one-tailed one-sample t-test, FDR corrected). For all panels, bars indicate mean effects and error bars indicate 95% C.I. All tests were based on n = 3 subjects, 160 odors/subject.

Extended Data Fig. 9 Dimensionality of encoded perceptual spaces for individual subjects.

a, Cumulative percentage of explained variance in the voxel-wise encoding weights as a function of the number of principal components, for individual subjects. b, Dimensionality parameter (κ) is proportional to area under the curve in a and reflects the number of principal components required to explain a given percentage of variance explained in each subject. Bars depict mean effect and error bars depict 95% C.I. (perc. bootstrap) across n = 3 subjects. The dimensionality of perceptual encoding is maximum in OFC in each subject and significantly different across areas (p = 0.000 (FWE corrected) against the null hypothesis κ(PirF) = κ(PirT) = κ(AMY) = κ(OFC), two-tailed bootstrap comparison, n = 3 subjects, 160 odors/subject). c, Dimensionality estimation adjusted for differences in ROI size. 25 voxels were chosen with replacement from each ROI to estimate the principal components in each bootstrap. d, Adjusted dimensionality increases from PirF to PirT to AMY and to OFC. Adjusted dimensionality is maximum in OFC and significantly different across areas (p = 0.002 (FWE corrected) against the null hypothesis κ(PirF) = κ(PirT) = κ(AMY) = κ(OFC), two-tailed bootstrap comparison, n = 3 subjects, 160 odors/subject). Error bars indicate 95% C.I. e Average PCA coefficients of perceptual feature weights for different principal components in PirF, PirT and AMY. PC1 is primarily driven by intensity, whereas subsequent components are more heterogeneous in all ROIs.

Extended Data Fig. 10 Subject-specific and cross-subject encoding model.

a, Mean prediction accuracy of encoding models based on fMRI data and perceptual ratings provided by the same subject (subject-specific encoding model [EM], dark) and fMRI data and ratings provided by different subjects (cross-subject EM, light bars). Subject-specific encoding models have a significantly higher prediction accuracy compared to cross-subject encoding models (F1,15 = 12.58, p = 0.016, repeated measures 2-way ANOVA with subjective-specific vs. cross-subject and ROI as factors). There was no significant main effect of ROI (F3,15 = 0.62, p = 0.615), and no significant interaction (F3,15 = 0.84, p = 0.494). b, Differences between the prediction accuracy of subject-specific and cross-subject encoding models. All encoding models were based on 14 principal components of perceptual ratings that explained at least 90% of variance. Lines depict individual subject pairs. Error bars are s.e.m. across all six subject pairs.

Supplementary information

Supplementary Information

Supplementary Tables 1 and 2.

Reporting Summary

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Sagar, V., Shanahan, L.K., Zelano, C.M. et al. High-precision mapping reveals the structure of odor coding in the human brain. Nat Neurosci 26, 1595–1602 (2023). https://doi.org/10.1038/s41593-023-01414-4

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1038/s41593-023-01414-4

  • Springer Nature America, Inc.

This article is cited by

Search

Navigation