Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Article
  • Published:

Rate and noise in human amygdala drive increased exploration in aversive learning

Nature (2025)Cite this article

Subjects

Abstract

To cope in uncertain environments, animals must balance their actions between using current resources and searching for new ones1. This exploration–exploitation dilemma has been studied extensively in paradigms involving positive outcomes, and neural correlates have been identified in frontal cortices and subcortical structures2,3,4,5,6,7,8,9,10,11, including the amygdala12. Importantly, exploration is just as essential for survival or well-being when trying to avoid negative outcomes, yet we do not know whether the single-neuron mechanisms that drive exploration are shared across positive and negative environments. Here we examined the dynamics of exploration when human participants engaged in a probabilistic learning task with intermixed loss and gain trials, while simultaneously recording single-neuron activity. We show that neurons of the amygdala and temporal cortex modulate their activity before a decision to explore in both loss and gain. Moreover, we find that humans exhibit more exploration when trying to avoid losses, and that an increase in the levels of noise in amygdala neurons contributes to this behaviour. Overall, we report that human exploration is driven by two distinct neural mechanisms, a valence-independent rate signal and a valence-dependent global noise signal. The results suggest a link between the heightened amygdala activity observed in mood disorders13,14 and higher exploration rates15,16,17 that underlie maladaptive and even pathological behaviours.

Access through your institution
Buy or subscribe

This is a preview of subscription content, access via your institution

Access options

Access through your institution

Buy this article

  • Purchase on SpringerLink
  • Instant access to full article PDF
Buy now

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Paradigm and learning behaviour in loss versus gain conditions.
Fig. 2: Higher exploration during loss trials.
Fig. 3: Neurons in the amygdala and the temporal cortex signal upcoming exploration.
Fig. 4: Rate signal for exploration is similar across the gain and loss conditions.
Fig. 5: Exploration in the loss condition is further driven by elevated noise in the amygdala.

Similar content being viewed by others

Data availability

The data reported in this study cannot be deposited in a public repository because they were derived from human patients and are subject to controlled access. To request access, please contact the lead authors: R.P. or I.S. The time frame for response is 12 months and conditioned on randomization of the data across participants, de-identification protocols and restrictions imposed by data use agreements between the requesting institute and Sourasky medical center.

Code availability

The code used for the analyses in this study is provided in the Supplementary Information.

References

  1. Cohen, J. D., McClure, S. M. & Yu, A. J. Should I stay or should I go? How the human brain manages the trade-off between exploitation and exploration. Phil. Trans. R. Soc. B 362, 933–942 (2007).

    Article  PubMed  PubMed Central  Google Scholar 

  2. Daw, N. D., O’Doherty, J. P., Dayan, P., Seymour, B. & Dolan, R. J. Cortical substrates for exploratory decisions in humans. Nature 441, 876–879 (2006).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  3. Blanchard, T. C. & Gershman, S. J. Pure correlates of exploration and exploitation in the human brain. Cogn. Affect. Behav. Neurosci. 18, 117–126 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  4. Hogeveen, J. et al. The neurocomputational bases of explore-exploit decision-making. Neuron 110, 1869–1879 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  5. Badre, D., Doll, B. B., Long, N. M. & Frank, M. J. Rostrolateral prefrontal cortex and individual differences in uncertainty-driven exploration. Neuron 73, 595–607 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  6. Gershman, S. J. Uncertainty and exploration. Decision 6, 277–286 (2019).

    Article  PubMed  Google Scholar 

  7. Trudel, N. et al. Polarity of uncertainty representation during exploration and exploitation in ventromedial prefrontal cortex. Nat. Hum. Behav. 5, 83–98 (2021).

    Article  PubMed  Google Scholar 

  8. Tomov, M. S., Truong, V. Q., Hundia, R. A. & Gershman, S. J. Dissociable neural correlates of uncertainty underlie different exploration strategies. Nat. Commun. 11, 2371 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  9. Zajkowski, W. K., Kossut, M. & Wilson, R. C. A causal role for right frontopolar cortex in directed, but not random, exploration. eLife 6, e27430 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  10. Pearson, J. M., Hayden, B. Y., Raghavachari, S. & Platt, M. L. Neurons in posterior cingulate cortex signal exploratory decisions in a dynamic multioption choice task. Curr. Biol. 19, 1532–1537 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  11. Barack, D. L., Chang, S. W. C. & Platt, M. L. Posterior cingulate neurons dynamically signal decisions to disengage during foraging. Neuron 96, 339–347 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  12. Costa, V. D., Mitz, A. R. & Averbeck, B. B. Subcortical substrates of explore-exploit decisions in primates. Neuron 103, 533–545 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  13. Alexandra Kredlow, M., Fenster, R. J., Laurent, E. S., Ressler, K. J. & Phelps, E. A. Prefrontal cortex, amygdala, and threat processing: implications for PTSD. Neuropsychopharmacology 47, 247–259 (2022).

    Article  CAS  PubMed  Google Scholar 

  14. Pitman, R. K. et al. Biological studies of post-traumatic stress disorder. Nat. Rev. Neurosci. 13, 769–787 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  15. Addicott, M. A., Pearson, J. M., Sweitzer, M. M., Barack, D. L. & Platt, M. L. A primer on foraging and the explore/exploit trade-off for psychiatry research. Neuropsychopharmacology 42, 1931–1939 (2017).

  16. Pulcu, E. & Browning, M. The misestimation of uncertainty in affective disorders. Trends Cogn. Sci. 23, 865–875 (2019).

    Article  PubMed  Google Scholar 

  17. Levy, I. & Schiller, D. Neural computations of threat. Trends Cogn. Sci. 25, 151–171 (2021).

    Article  PubMed  Google Scholar 

  18. Ebitz, R. B., Albarran, E. & Moore, T. Exploration disrupts choice-predictive signals and alters dynamics in prefrontal cortex. Neuron 97, 475 (2018).

    Article  CAS  PubMed  Google Scholar 

  19. Costa, V. D. & Averbeck, B. B. Primate orbitofrontal cortex codes information relevant for managing explore–exploit tradeoffs. J. Neurosci. 40, 2553–2561 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Ogasawara, T. et al. A primate temporal cortex-zona incerta pathway for novelty seeking. Nat. Neurosci. 25, 50–60 (2022).

    Article  CAS  PubMed  Google Scholar 

  21. Aquino, T. G., Cockburn, J., Mamelak, A. N., Rutishauser, U. & O’Doherty, J. P. Neurons in human pre-supplementary motor area encode key computations for value-based choice. Nat. Hum. Behav. 7, 970–985 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  22. Domenech, P., Rheims, S. & Koechlin, E. Neural mechanisms resolving exploitation-exploration dilemmas in the medial prefrontal cortex. Science 369, eabb0184 (2020).

  23. Aquino, T. G., Courellis, H., Mamelak, A. N., Rutishauser, U. & O′Doherty, J. P. Encoding of predictive associations in human prefrontal and medial temporal neurons during pavlovian appetitive conditioning. J. Neurosci. 44, e1628232024 (2024).

  24. Barberini, C. L., Morrison, S. E., Saez, A., Lau, B. & Salzman, C. D. Complexity and competition in appetitive and aversive neural circuits. Front. Neurosci. 6, 170 (2012).

  25. Herry, C. & Johansen, J. P. Encoding of fear learning and memory in distributed neuronal circuits. Nat. Neurosci. 17, 1644–1654 (2014).

    Article  CAS  PubMed  Google Scholar 

  26. Janak, P. H. & Tye, K. M. From circuits to behaviour in the amygdala. Nature 517, 284–292 (2015).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  27. Pape, H. C. & Pare, D. Plastic synaptic networks of the amygdala for the acquisition, expression, and extinction of conditioned fear. Physiol. Rev. 90, 419–463 (2010).

    Article  CAS  PubMed  Google Scholar 

  28. Blanco, N. J., Otto, A. R., Maddox, W. T., Beevers, C. G. & Love, B. C. The influence of depression symptoms on exploratory decision-making. Cognition 129, 563–568 (2013).

    Article  PubMed  Google Scholar 

  29. de Berker, A. O. et al. Computations of uncertainty mediate acute stress responses in humans. Nat. Commun. 7, 10996 (2016).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

  30. Browning, M., Behrens, T. E., Jocham, G., O’Reilly, J. X. & Bishop, S. J. Anxious individuals have difficulty learning the causal statistics of aversive environments. Nat. Neurosci. 18, 590–596 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  31. Aberg, K. C., Toren, I. & Paz, R. A neural and behavioral trade-off between value and uncertainty underlies exploratory decisions in normative anxiety. Mol. Psychiatry 27, 1573–1587 (2022).

    Article  PubMed  Google Scholar 

  32. Fan, H., Gershman, S. J. & Phelps, E. A. Trait somatic anxiety is associated with reduced directed exploration and underestimation of uncertainty. Nat. Hum. Behav. 7, 102–113 (2023).

    Article  PubMed  Google Scholar 

  33. Aberg, K. C. & Paz, R. The neurobehavioral correlates of exploration without learning: trading off value for explicit, prospective, and variable information gains. Cell Rep. 43, 113880 (2024).

    Article  CAS  PubMed  Google Scholar 

  34. Carleton, R. N. Into the unknown: a review and synthesis of contemporary models involving uncertainty. J. Anxiety Disord. 39, 30–43 (2016).

    Article  PubMed  Google Scholar 

  35. Ölveczky, B. P., Andalman, A. S. & Fee, M. S. Vocal experimentation in the juvenile songbird requires a basal ganglia circuit. PLoS Biol. 3, e153 (2005).

    Article  PubMed  PubMed Central  Google Scholar 

  36. Mandelblat-Cerf, Y., Paz, R. & Vaadia, E. Trial-to-trial variability of single cells in motor cortices is dynamically modified during visuomotor adaptation. J. Neurosci. 29, 15053–15062 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  37. Maimon, G. & Assad, J. A. Beyond Poisson: increased spike-time regularity across primate parietal cortex. Neuron 62, 426–440 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Stein, R. B., Gossen, E. R. & Jones, K. E. Neuronal variability: noise or part of the signal? Nat. Rev. Neurosci. 6, 389–397 (2005).

    Article  CAS  PubMed  Google Scholar 

  39. Faisal, A. A., Selen, L. P. & Wolpert, D. M. Noise in the nervous system. Nat. Rev. Neurosci. 9, 292–303 (2008).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  40. Echeveste, R. & Lengyel, M. The redemption of noise: inference with neural populations. Trends Neurosci. 41, 767–770 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  41. Krueger, P. M., Wilson, R. C. & Cohen, J. D. Strategies for exploration in the domain of losses. Judgm. Decis. Making 12, 104–117 (2017).

    Article  Google Scholar 

  42. Lejarraga, T. & Hertwig, R. How the threat of losses makes people explore more than the promise of gains. Psychon. Bull. Rev. 24, 708–720 (2017).

    Article  PubMed  Google Scholar 

  43. Seymour, B., Maruyama, M. & De Martino, B. When is a loss a loss? Excitatory and inhibitory processes in loss-related decision-making. Curr. Opin. Behav. Sci. 5, 122–127 (2015).

    Article  Google Scholar 

  44. Amaya, K. A. & Smith, K. S. Neurobiology of habit formation. Curr. Opin. Behav. Sci. 20, 145–152 (2018).

    Article  Google Scholar 

  45. Cockburn, J., Man, V., Cunningham, W. A. & O’Doherty, J. P. Novelty and uncertainty regulate the balance between exploration and exploitation through distinct mechanisms in the human brain. Neuron 110, 2691–2702 (2022).

    Article  CAS  PubMed  Google Scholar 

  46. Unruh-Pinheiro, A. et al. Single-neuron correlates of decision confidence in the human medial temporal lobe. Curr. Biol. 30, 4722–4732 (2020).

    Article  CAS  PubMed  Google Scholar 

  47. Mormann, F., Bausch, M., Knieling, S. & Fried, I. Neurons in the human left amygdala automatically encode subjective value irrespective of task. Cereb. Cortex 29, 265–272 (2019).

    Article  CAS  PubMed  Google Scholar 

  48. Aquino, T. G. et al. Value-related neuronal responses in the human amygdala during observational learning. J. Neurosci. 40, 4761–4772 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  49. Kehl, M. S. et al. Single-neuron representations of odours in the human brain. Nature 634, 626–634 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  50. Reitich-Stolero, T. et al. Aversive generalization in human amygdala neurons. Curr. Biol. 35, 1137–1144 (2025).

    Article  CAS  PubMed  Google Scholar 

  51. Madarasz, T. J. et al. Evaluation of ambiguous associations in the amygdala by learning the structure of the environment. Nat. Neurosci. 19, 965–972 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  52. Herry, C. et al. Processing of temporal unpredictability in human and animal amygdala. J. Neurosci. 27, 5958–5966 (2007).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  53. Livneh, U. & Paz, R. Amygdala-prefrontal synchronization underlies resistance to extinction of aversive memories. Neuron 75, 133–142 (2012).

    Article  CAS  PubMed  Google Scholar 

  54. Monosov, I. E. How outcome uncertainty mediates attention, learning, and decision-making. Trends Neurosci. 43, 795–809 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  55. Jezzini, A., Bromberg-Martin, E. S., Trambaiolli, L. R., Haber, S. N. & Monosov, I. E. A prefrontal network integrates preferences for advance information about uncertain rewards and punishments. Neuron 109, 2339–2352 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  56. Kolling, N. et al. Value, search, persistence and model updating in anterior cingulate cortex. Nat. Neurosci. 19, 1280–1285 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  57. Kobayashi, K. & Kable, J. W. Neural mechanisms of information seeking. Neuron 112, 1741–1756 (2024).

    Article  CAS  PubMed  Google Scholar 

  58. Litwin-Kumar, A. & Doiron, B. Slow dynamics and high variability in balanced cortical networks with clustered connections. Nat. Neurosci. 15, 1498–1505 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  59. Shadlen, M. N. & Newsome, W. T. The variable discharge of cortical neurons: implications for connectivity, computation, and information coding. J. Neurosci. 18, 3870–3896 (1998).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  60. Rubin, R., Abbott, L. F. & Sompolinsky, H. Balanced excitation and inhibition are required for high-capacity, noise-robust neuronal selectivity. Proc. Natl Acad. Sci. USA 114, E9366–E9375 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  61. Levy, D. R. et al. Dynamics of social representation in the mouse prefrontal cortex. Nat. Neurosci. 22, 2013–2022 (2019).

    Article  CAS  PubMed  Google Scholar 

  62. Aston-Jones, G. & Cohen, J. D. An integrative theory of locus coeruleus-norepinephrine function: adaptive gain and optimal performance. Ann. Rev. Neurosci. 28, 403–450 (2005).

    Article  CAS  PubMed  Google Scholar 

  63. Payzan-LeNestour, E., Dunne, S., Bossaerts, P. & O’Doherty, J. P. The neural representation of unexpected uncertainty during value-based decision making. Neuron 79, 191–201 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  64. Feng, Y. Y., Bromberg-Martin, E. S. & Monosov, I. E. Dorsal raphe neurons integrate the values of reward amount, delay, and uncertainty in multi-attribute decision-making. Cell Rep. 43, 114341 (2024).

    Article  CAS  PubMed  Google Scholar 

  65. Grossman, C. D., Bari, B. A. & Cohen, J. Y. Serotonin neurons modulate learning rate through uncertainty. Curr. Biol. 32, 586–599 (2022).

    Article  CAS  PubMed  Google Scholar 

  66. Lottem, E. et al. Activation of serotonin neurons promotes active persistence in a probabilistic foraging task. Nat. Commun. 9, 1000 (2018).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

  67. Wilson, R. C., Geana, A., White, J. M., Ludvig, E. A. & Cohen, J. D. Humans use directed and random exploration to solve the explore-exploit dilemma. J. Exp. Psychol. Gen. 143, 2074–2081 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

  68. Aberg, K. C. & Paz, R. Stress-induced avoidance in mood disorders. Nat. Hum. Behav. 6, 915–918 (2022).

    Article  PubMed  Google Scholar 

  69. Kahneman, D. & Tversky, A. Prospect theory: an analysis of decision under risk. Econometrica 47, 263 (1979).

    Article  MathSciNet  MATH  Google Scholar 

  70. Misra, A. et al. Methods for implantation of micro-wire bundles and optimization of single/multi-unit recordings from human mesial temporal lobe. J. Neural Eng. 11, 026013 (2014).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  71. Fried, I. et al. Cerebral microdialysis combined with single-neuron and electroencephalographic recording in neurosurgical patients. J. Neurosurg. 91, 697–705 (1999).

    Article  CAS  PubMed  Google Scholar 

  72. Amunts, K., Mohlberg, H., Bludau, S. & Zilles, K. Julich-Brain: a 3D probabilistic atlas of the human brain’s cytoarchitecture. Science 369, 988–992 (2020).

    Article  ADS  CAS  PubMed  Google Scholar 

  73. Rutishauser, U. et al. Representation of retrieval confidence by single neurons in the human medial temporal lobe. Nat. Neurosci. 18, 1041–1050 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  74. Kamiński, J., Brzezicka, A., Mamelak, A. N. & Rutishauser, U. Combined phase-rate coding by persistently active neurons as a mechanism for maintaining multiple items in working memory in humans. Neuron 106, 256–264 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  75. Joshua, M., Elias, S., Levine, O. & Bergman, H. Quantifying the isolation quality of extracellularly recorded action potentials. J. Neurosci. Methods 163, 267–282 (2007).

    Article  PubMed  Google Scholar 

  76. Reitich-Stolero, T. & Paz, R. Affective memory rehearsal with temporal sequences in amygdala neurons. Nat. Neurosci. 22, 2050–2059 (2019).

    Article  CAS  PubMed  Google Scholar 

  77. Gershman, S. J. Deconstructing the human algorithms for exploration. Cognition 173, 34–42 (2018).

    Article  PubMed  Google Scholar 

  78. Frank, M. J., Doll, B. B., Oas-Terpstra, J. & Moreno, F. Prefrontal and striatal dopaminergic genes predict individual differences in exploration and exploitation. Nat. Neurosci. 12, 1062–1068 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  79. Piray, P., Dezfouli, A., Heskes, T., Frank, M. J. & Daw, N. D. Hierarchical Bayesian inference for concurrent model fitting and comparison for group studies. PLoS Comput. Biol. 15, e1007043 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

  80. Rigoux, L., Stephan, K. E., Friston, K. J. & Daunizeau, J. Bayesian model selection for group studies—revisited. Neuroimage 84, 971–985 (2014).

    Article  CAS  PubMed  Google Scholar 

  81. Wilson, R. C. & Collins, A. G. Ten simple rules for the computational modeling of behavioral data. eLife 8, e49547 (2019).

  82. Shadlen, M. N., Hanks, T. D., Churchland, A. K., Kiani, R. & Yang, T. in Bayesian Brain: Probabilistic Approaches to Neural Coding (eds Doya, K. et al.) 209–237 (MIT Press, 2006).

  83. Ison, M. J. et al. Selectivity of pyramidal cells and interneurons in the human medial temporal lobe. J. Neurophysiol. 106, 1713–1721 (2011).

    Article  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

The work was supported by ERC-2023-ADG 101142391 and ISF 1467/24 grants to R. Paz. We thank L. Bergman for help in electrode placement and S. Devore for critical reading and comments on the manuscript.

Author information

Author notes

  1. These authors contributed equally: Ido Strauss, Rony Paz

Authors and Affiliations

  1. Department of Brain Sciences, Weizmann Institute of Science, Rehovot, Israel

    Tamar Reitich-Stolero, Kristoffer C. Aberg, Dean Halperin, Carmel Ariel & Rony Paz

  2. Azrieli Institute for Brain and Neural Sciences, Weizmann Institute of Science, Rehovot, Israel

    Kristoffer C. Aberg & Rony Paz

  3. Department of Neurology and Neurosurgery, School of Medicine, Tel Aviv University, Tel Aviv, Israel

    Genela Morris & Ido Strauss

  4. Epilepsy Unit, Neurological Institute, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel

    Lilach Goldstein & Firas Fahoum

  5. Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel

    Lilach Goldstein & Firas Fahoum

  6. Functional Neurosurgery Unit, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel

    Ido Strauss

Authors
  1. Tamar Reitich-Stolero
    View author publications

    Search author on:PubMed Google Scholar

  2. Kristoffer C. Aberg
    View author publications

    Search author on:PubMed Google Scholar

  3. Dean Halperin
    View author publications

    Search author on:PubMed Google Scholar

  4. Carmel Ariel
    View author publications

    Search author on:PubMed Google Scholar

  5. Genela Morris
    View author publications

    Search author on:PubMed Google Scholar

  6. Lilach Goldstein
    View author publications

    Search author on:PubMed Google Scholar

  7. Firas Fahoum
    View author publications

    Search author on:PubMed Google Scholar

  8. Ido Strauss
    View author publications

    Search author on:PubMed Google Scholar

  9. Rony Paz
    View author publications

    Search author on:PubMed Google Scholar

Contributions

T.R.S. and R.P. designed all experiments. T.R.S. conducted the behavioural pilot experiments. T.R.S., I.S., F.F., C.A., L.G. and G.M. conducted the neural recordings experiments. T.R.S. analysed the behavioural and the neural data and performed neural modelling. K.C.A. performed behavioural modelling and contributed to analyses of behaviour. D.H. contributed to neural data analyses. C.A. conducted the mixed block behavioural experiment. K.C.A. and D.H. contributed to writing the paper. T.R.S. and R.P. wrote the paper.

Corresponding authors

Correspondence to Tamar Reitich-Stolero or Rony Paz.

Ethics declarations

Competing interests

The authors declare no competing interests.

Peer review

Peer review information

Nature thanks Ilya Monosov, Florian Mormann and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

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

Extended data figures and tables

Extended Data Fig. 1 Behavioural results in a pilot study validate the higher exploration in loss.

(a) Proportion of correct choices from four single participants in the pilot behavioural study (healthy participants, n = 46), in gain trials (blue) and loss trials (orange). Single trial-by-trial correct and incorrect choices are shown as dots in the top and bottom, respectively. (b) Proportion of correct choices for participants in the pilot study for gain trials (blue) and loss trials (orange). Shown for all participants (left, n = 46) and for participants exhibiting significant learning (right, n = 38). There was a significant performance increase between early and late trials of the gain and of the loss condition (middle inset, sign rank test comparing the first and last 20 trial in gain: Z = −2.85, p < 0.004; in loss: Z = −4.06, p < 10−4). Individual participant values are shown as grey lines. (c) Proportion of exploration choices as a function of trial (see main text and Fig. 2a for criterion) for all participants (left) and for participants exhibiting significant learning (right). Participants explored significantly more in the loss condition (middle inset, sign rank test on the proportion of exploration, n = 46, Z = −5.8, p < 10−8). Individual participant values are shown as grey lines.

Extended Data Fig. 2 Reaction times (RT) are higher in loss and for exploration choices.

(a) Reaction times in gain (blue) and loss (orange) choices. Left: kernel density estimates of RT across all participants and choices. The black line represents the median and the box represents the middle 50% of the distribution (25–75%). Right top: Average RT for each participant across all gain choices (x axis) is significantly smaller than in loss choices (y axis, sign rank test, n = 22, Z = −4.01, p < 10−4). Right bottom: RT as a function of trial during learning, averaged across participants (10 trials running average). (b) Histogram of RT in single trials for exploration and exploitation choices (red/green). Full and dashed lines mark the median and mean presentation time, respectively. (c) The difference in average RT between loss and gain trials is positively correlated with the difference in the proportion of exploration (n = 22, r = 0.65 p = 0.001), suggesting that participants with longer loss-related RT compared to gain-related RT also explored more in loss compared to gain. (d) Average RT for participants in exploration versus exploitation choices in gain and loss (left), and distributions of RT of exploration/exploitation choices in gain and loss trials across all participants (right). There was a significant interaction between EE and valence (n = 3262, F(1,3258) = 13.23, p < 10−3), with a larger RT in gain-exploration compared to gain-exploitation (n = 1631, t1629 = 6.6, p < 10−10, dcohen = 0.67) and no difference between loss-exploration and loss-exploitation (n = 1631, t1629 = 0.65, p > 0.5).

Extended Data Fig. 3 The EE signal starts to increase already after the presentation of the shapes.

Same as in Fig. 3b,e,f, but with neural activity aligned to the appearance of the shapes or tone. (a) Proportion of neurons with a significant effect for exploration and exploitation trials in the amygdala and temporal cortex (ANOVA model, accounting for EE, choice identity and the sign of the prediction error). The full and dashed red lines mark the median and mean reaction-times, respectively. Black dashed line depicts chance level. (b) The difference between explained variability (R2) in models with and without the EE predictor. (c) Effect size (explained variance) of FR by EE classification, after accounting for PE-sign and choice identity. (d) Mean and standard error (SE) of the normalized FR across all amygdala and temporal cortex neurons in exploration and exploitation trials (red and green, respectively). The times of significant difference are marked by black dots (two-tailed t-test over all single neurons, FDR corrected). (e) Same as in (a,d) but with neural activity aligned to the appearance of the tone that indicates the trial valence (loss/gain). The lack of signal serves as a control to the main finding that FR modulation concerns the choice, as it appears after the shapes (A-D) and further increases towards the choice action (main Fig. 3).

Extended Data Fig. 4 EE signal in hippocampal regions (n = 88), frontal- (n = 109), and sensory-motor (n = 63) cortices.

(a) Mean and SE of the normalized FR across neurons in exploration and exploitation trials (red and green, respectively) as a function of the time from the choice. While there seems to be an exploration signal in hippocampal regions, it was not significant during the pre-choice epoch, and it was not correlated with behaviour (see main text). (b) Proportion of units with a significant effect for exploration and exploitation trials (as in Fig. 3e, ANOVA model, accounting for EE, choice identity and the sign of the prediction error). The proportion in the hippocampus was not significantly above chance (FDR corrected as in Fig. 3e). (c) The difference between explained variability (R2) in models with and without the EE predictor.

Extended Data Fig. 5 Noise levels between loss and gain.

(a-b) Similar data and presentation as in Fig. 5b,e, but shown here is the median (instead of the average). (a) CV ratio between loss and gain \(\left(\frac{C{V}_{{loss}}}{C{V}_{{gain}}}\right)\) aligned on the choice. (b) CV ratio between loss and gain \(\left(\frac{C{V}_{{loss}}}{C{V}_{{gain}}}\right)\) aligned on tone presentation. (c) The variability of the inter-spike-intervals (ISI) distributions after the appearance of shapes are higher in loss compared to gain. ISIs were taken from a one second window after the shapes (left, n = 43, tdf=42 = 1.8, p = 0.04) and after the tone (right, n = 43, tdf=42 = 1.27, p = 0.1).

Extended Data Fig. 6 Firing rates in amygdala neurons do not explain the higher exploration in loss.

(a) Firing rates (FR) in loss compared to gain in the two amygdala neurons that are shown in Fig. 5a. (b) FR ratio between loss and gain \(\left(\frac{F{R}_{{loss}}}{F{R}_{{gain}}}\right)\) averaged over all non-EE amygdala neurons (n = 35), shown as the ratio between the FRs averaged over one second pre-choice for every neuron (top), and over time from choice (bottom). (c) FR across all gain trials and loss trials, averaged over all non-EE amygdala neurons (top, n = 35) and all EE amygdala neurons (bottom, n = 8). The pre-choice FR was not significantly different (t-test for each window). (d) Pre-choice FR of amygdala neurons was not correlated with probability of exploration in loss trials, across any window size (inset). (e) The correlation between CV and probability of exploration after controlling for the FR (partial correlation), presented for different window sizes (full circles are significant at p < 0.05). For comparison, the bottom panel shows the correlation without controlling for FR (same panel as in Fig. 5j, inset).

Extended Data Fig. 7 Exploration by elevated noise is enhanced in the amygdala compared to other brain regions.

The same measures as in Fig. 5 are plotted for the other brain regions. The left column (amygdala) are the same panels as in Fig. 5, shown here again for comparison. (a) Ratio between the raw CV in the loss and gain conditions \(\left(\frac{C{V}_{{loss}}}{C{V}_{{gain}}}\right)\) as a function of time from the choice, averaged over all neurons. (b) CV across all gain trials and loss trials averaged over all neurons in a region. (c) Standard deviation of the inter-spike-interval (ISI) distribution across all trials, at one second pre-choice, and for each neuron in loss versus gain (Temporal: \({t}_{{df}=76}=0.68,p > 0.05\); Prefrontal:: \({t}_{{df}=107}=2.53,p < 0.01\); Hippocampal: \({t}_{{df}=84}=0.3,p > 0.05;\) Sensory motor: \({t}_{{df}=62}=0.68,p > 0.05\)). (d) Pre-choice CV in loss is not correlated with probability of exploration, across most window sizes (bottom inset). Notice that the frontal cortex also shows elevated noise in loss trials, but noise levels are not correlated with behavioural exploration.

Extended Data Fig. 8 Neural noise links uncertainty to exploration.

All figures show the correlation between residual values of two variables after accounting for a third one, i.e. the partial correlation. The upper row uses the total uncertainty (TUt) calculated from the model, and the bottom row uses trial number (because uncertainty decreases with trial number, we use it as another validation). (a) Pre-choice CV increases with uncertainty (top) and decreases with trial number (bottom), even after accounting for exploration. Inset: the partial correlation is robust across window size (filled circle: p < 0.05). This suggests that pre-choice CV increases with uncertainty independent of exploration. (b) Exploration increases with pre-choice CV after accounting for uncertainty (top) as well as after accounting for trial number (bottom). Inset: the partial correlation is robust across window size (filled circle: p < 0.05). This suggests that exploration increases with noise independent of uncertainty. (c) Exploration is not correlated with uncertainty (top) and not with trial number (bottom) after accounting for pre-choice CV. The lack of correlation suggests that there is no direct relationship between uncertainty and exploration, and that the CV mediates the two.

Extended Data Fig. 9 Decision-making models validate the combined effect of noise and rate signal.

For both models, we postulated multiple normal subpopulation (a) or multiple DDM neurons (b) converging to a “post-synaptic” neuron/region that is activated by either of the inputs (see methods, [equation 2]). (a) An illustration of the ‘normal’ model: example distributions of the sum of FR of k identically and independently distributed (i.i.d) neurons with similar mean FR and high standard deviation (σL, orange) or low standard deviation (σG, blue). The probability of random crossing of a threshold (dashed black line) under the noisier distribution (high standard deviation, σL) is larger than under the less noisy distribution (quantified by the red and green filled areas, respectively). See methods under normal model. (b) An illustration of the DDM model (with μ = 1, σ = 2.3, see methods). Crossing the positive boundary (green and yellow traces) was modelled as exploitation and crossing the negative one (orange trace) was modelled as exploration. (c) The probability of exploration increases with the noise level (σ) in both models. (d) The ratio of the noise levels (\(\frac{{\sigma }_{L}}{{\sigma }_{G}}\), y axis, see methods [equations 1,3]) that is associated with the average probability of exploration experimentally observed in the loss \(({p}_{{loss}}=0.25)\) and gain \(({p}_{{gain}}=0.06)\) as a function of n subpopulations (Gaussian model) or n neurons (DDM). The ratio decreases and reaches a plateau with n. Notice that in this framework and for equal mean exploitation signal, the ratio between standard deviations of models with equal probabilities only depends on the probabilities in each condition. Similarly, the ratio between EE signals can be described in terms of the probabilities and standard deviations in each condition. (e) Noise-level ratio \(\left(\frac{{\sigma }_{L}}{{\sigma }_{G}}{colormap}\right)\) for different exploration rates in the loss and gain conditions, assuming equal signal strength and a large enough subpopulation. White dots mark the actual choice probabilities of single participants. White cross marks the mean and SE across our participants, overlaid by 2D kernel density contours. Black dots mark the expected choice probabilities for the experimentally measured noise ratio in amygdala neurons (the 75th percentile, see methods). Shown for both models (DDM also shown in main Fig. 5l). The noise-level \(\left(\frac{{\sigma }_{L}}{{\sigma }_{G}}\right)\) is computed from the CV. (f) Same as (e) but with the noise-level \(\left(\frac{{\sigma }_{L}}{{\sigma }_{G}}\right)\) computed from the std(ISI) instead of the CV. Notice that in all four cases (e,f), the participants’ mean behaviour (white cross) closely matches the expected behaviour from the models when using the noise levels measured in amygdala neurons (black dots).

Extended Data Fig. 10 Loss aversion does not account for higher exploration in loss.

(a,b) A pilot behavioural study with balanced value, namely without correction for loss aversion as implemented in the main study (n = 26, gain: +5, loss: −5). (a) Proportion of exploration choices as a function of trial (similar presentation and analyses as in main Fig. 1 and Extended Data Fig. 1). (b) Left: exploration term (inverse temperature, β). Inset: single participant exploration terms (mean and SE) were lower in loss, suggesting higher exploration (Left-tailed sign-rank test, Z = −4.46, p < 10−5). Right: learning rate (LR, α). Inset: single participant learning rates were actually higher in loss (Two-tailed sign-rank test, Z = 3.42, p < 10−3). (c,d,e,f) A loss-aversion task was performed for part of the participants in the pilot behavioural study described in Extended Data Fig. 1 (n = 18). (c) Three participants’ decision matrix, showing accepted gambles in white and rejected ones in black (top row), together with the predicted probabilities by the fitted logistic loss aversion function (bottom row). (d) Average decision matrix across all participants. Equal values diagonal is marked in blue asterisks (no loss aversion) and the value (Lambda = 2) used for correction in the main experiment is marked in red (notice the different values on the x-axis and y-axis). (e) Distribution of individual loss aversion index for all participants (Median is marked in dashed line). (f) There was no significant correlation between loss aversion index and the proportion of exploration in loss trials (Main, r = 0.01, p = 0.95), gain trials (right inset, r = 0.03, p = 0.9) or the difference between them (left inset, r = −0.16, p = 0.52). (g) A mixed block behavioural experiment with a similar design. Each participant performed four blocks with intermixed trials of pairs with different outcomes. Each bar represents a single trial type with similar probabilities of feedback as in the main experiment. The valence and magnitude of potential outcomes appear below (only the valenced outcomes are shown; a null outcome was also possible). The proportion of exploration was determined mainly by valence (c.f. orange versus blue bars) and not by the magnitude or the paired alternative: it was always higher for loss trials (orange bars) than for gain trials (blue bars), even when the two loss outcomes in a pair were of different loss magnitudes (i.e. −5 and −10). Moreover, notice that participant explored more to avoid a potential loss of 10 points compared to exploration when trying to gain 10 points, for all gain blocks, and that participants explored less when trying to gain 5 points compared to exploration when trying to avoid loss of 5 points, for all loss blocks. Accordingly, it is the valence of potential outcomes, rather than their magnitude, that drives exploration (Two-way repeated measures ANOVA; Loss vs gain: F = 49.94, p < 0.001, η2 = 0.52; Magnitude: F = 1.961, p = 0.17, η2 = 0.04; Interaction: F = 4.47, p = 0.04, η2 = 0.09). ** p < 0.01, *** p < 0.001, all pass Bonferroni correction for multiple comparisons.

Supplementary information

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Reitich-Stolero, T., Aberg, K.C., Halperin, D. et al. Rate and noise in human amygdala drive increased exploration in aversive learning. Nature (2025). https://doi.org/10.1038/s41586-025-09466-1

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1038/s41586-025-09466-1

Search

Advanced search

Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing