By combining computational, morphological, and functional analyses, this study relates latent markers of associative threat learning to overt post-traumatic stress disorder (PTSD) symptoms in combat veterans. Using reversal learning, we found that symptomatic veterans showed greater physiological adjustment to cues that did not predict what they had expected, indicating greater sensitivity to prediction errors for negative outcomes. This exaggerated weighting of prediction errors shapes the dynamic learning rate (associability) and value of threat predictive cues. The degree to which the striatum tracked the associability partially mediated the positive correlation between prediction-error weights and PTSD symptoms, suggesting that both increased prediction-error weights and decreased striatal tracking of associability independently contribute to PTSD symptoms. Furthermore, decreased neural tracking of value in the amygdala, in addition to smaller amygdala volume, independently corresponded to higher PTSD symptom severity. These results provide evidence for distinct neurocomputational contributions to PTSD symptoms.

Access optionsAccess options

Rent or Buy article

Get time limited or full article access on ReadCube.


All prices are NET prices.

Code availability

The code used for the analyses is available online at: http://osf.io/rxsw2/.

Data availability

Data used to support the conclusions of this study is available online at: http://osf.io/rxsw2/.

Additional information

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

Change history

  • 13 February 2019

    The original and corrected figures are shown in the accompanying Author Correction.


  1. 1.

    American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, 4th edn (American Psychiatric Publishing, Arlington, 2013).

  2. 2.

    Pietrzak, R. H., Tsai, J., Harpaz-Rotem, I., Whealin, J. M. & Southwick, S. M. Support for a novel five-factor model of posttraumatic stress symptoms in three independent samples of Iraq/Afghanistan veterans: a confirmatory factor analytic study. J. Psychiatr. Res. 46, 317–322 (2012).

  3. 3.

    Harpaz-Rotem, I., Tsai, J., Pietrzak, R. H. & Hoff, R. The dimensional structure of posttraumatic stress symptomatology in 323,903 U.S. veterans. J. Psychiatr. Res. 49, 31–36 (2014).

  4. 4.

    Lissek, S. & van Meurs, B. Learning models of PTSD: theoretical accounts and psychobiological evidence. Int. J. Psychophysiol. 98, 594–605 (2015).

  5. 5.

    Pavlov, I. Conditioned Reflexes (Oxford Univ. Press, Oxford, 1927).

  6. 6.

    Pearce, J. M. & Hall, G. A model for Pavlovian learning: variations in the effectiveness of conditioned but not of unconditioned stimuli. Psychol. Rev. 87, 532–552 (1980).

  7. 7.

    Schiller, D., Levy, I., Niv, Y., LeDoux, J. E. & Phelps, E. A. From fear to safety and back: reversal of fear in the human brain. J. Neurosci. 28, 11517–11525 (2008).

  8. 8.

    Li, J., Schiller, D., Schoenbaum, G., Phelps, E. A. & Daw, N. D. Differential roles of human striatum and amygdala in associative learning. Nat. Neurosci. 14, 1250–1252 (2011).

  9. 9.

    Atlas, L. Y., Doll, B. B., Li, J., Daw, N. D. & Phelps, E. A. Instructed knowledge shapes feedback-driven aversive learning in striatum and orbitofrontal cortex, but not the amygdala. eLife 5, e15192 (2016).

  10. 10.

    Duits, P. et al. Updated meta-analysis of classical fear conditioning in the anxiety disorders. Depress. Anxiety 32, 239–253 (2015).

  11. 11.

    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).

  12. 12.

    LeDoux, J. E. Emotion circuits in the brain. Annu. Rev. Neurosci. 23, 155–184 (2000).

  13. 13.

    Pietrzak, R. H. et al. Amygdala-hippocampal volume and the phenotypic heterogeneity of posttraumatic stress disorder: a cross-sectional study. JAMA Psychiatry 72, 396–398 (2015).

  14. 14.

    Admon, R. et al. Human vulnerability to stress depends on amygdala’s predisposition and hippocampal plasticity. Proc. Natl Acad. Sci. USA 106, 14120–14125 (2009).

  15. 15.

    Neumeister, P. et al. Specific amygdala response to masked fearful faces in post-traumatic stress relative to other anxiety disorders. Psychol. Med. 48, 1209–1217 (2018).

  16. 16.

    Rescorla, R. & Wagner, A. in Classical Conditioning II: Current Research and Theory (eds Black A. H. & Prokasy, W. F.) (Appleton-Century-Crofts, New York, 1972).

  17. 17.

    Roesch, M. R., Esber, G. R., Li, J., Daw, N. D. & Schoenbaum, G. Surprise! Neural correlates of Pearce–Hall and Rescorla–Wagner coexist within the brain. Eur. J. Neurosci. 35, 1190–1200 (2012).

  18. 18.

    Jin, J., Zelano, C., Gottfried, J. A. & Mohanty, A. Human amygdala represents the complete spectrum of subjective valence. J. Neurosci. 35, 15145–15156 (2015).

  19. 19.

    Belova, M. A., Paton, J. J., Morrison, S. E. & Salzman, C. D. Expectation modulates neural responses to pleasant and aversive stimuli in primate amygdala. Neuron 55, 970–984 (2007).

  20. 20.

    Klavir, O., Genud-Gabai, R. & Paz, R. Functional connectivity between amygdala and cingulate cortex for adaptive aversive learning. Neuron 80, 1290–1300 (2013).

  21. 21.

    Genud-Gabai, R., Klavir, O. & Paz, R. Safety signals in the primate amygdala. J. Neurosci. 33, 17986–17994 (2013).

  22. 22.

    Morey, R. A. et al. Mid-Atlantic MIRECC Workgroup. Amygdala volume changes in posttraumatic stress disorder in a large case-controlled veterans group. Arch. Gen. Psychiatry 69, 1169–1178 (2012).

  23. 23.

    Wrocklage, K. M. et al. Cortical thickness reduction in combat exposed U.S. veterans with and without PTSD. Eur. Neuropsychopharmacol. 27, 515–525 (2017).

  24. 24.

    Raio, C. M., Hartley, C. A., Orederu, T. A., Li, J. & Phelps, E. A. Stress attenuates the flexible updating of aversive value. Proc. Natl Acad. Sci. USA 114, 11241–11246 (2017).

  25. 25.

    Roesch, M. R., Calu, D. J., Esber, G. R. & Schoenbaum, G. Neural correlates of variations in event processing during learning in basolateral amygdala. J. Neurosci. 30, 2464–2471 (2010).

  26. 26.

    Schultz, W. Dopamine neurons and their role in reward mechanisms. Curr. Opin. Neurobiol. 7, 191–197 (1997).

  27. 27.

    O’Doherty, J. P., Dayan, P., Friston, K., Critchley, H. & Dolan, R. J. Temporal difference models and reward-related learning in the human brain. Neuron 38, 329–337 (2003).

  28. 28.

    Preuschoff, K. & Bossaerts, P. Adding prediction risk to the theory of reward learning. Ann. N.Y. Acad. Sci. 1104, 135–146 (2007).

  29. 29.

    Behrens, T. E., Woolrich, M. W., Walton, M. E. & Rushworth, M. F. Learning the value of information in an uncertain world. Nat. Neurosci. 10, 1214–1221 (2007).

  30. 30.

    Etkin, A. & Wager, T. D. Functional neuroimaging of anxiety: a meta-analysis of emotional processing in PTSD, social anxiety disorder, and specific phobia. Am. J. Psychiatry 164, 1476–1488 (2007).

  31. 31.

    Ruderman, L. et al. Posttraumatic symptoms and aversion to ambiguous losses in combat veterans. Depress. Anxiety 33, 606–613 (2016).

  32. 32.

    Brown, V. M. et al. Associability-modulated loss learning is increased in post-traumatic stress disorder. eLife 7, e30150 (2018).

  33. 33.

    Daw, N. D. in Decision Making, Affect, and Learning: Attention and Performance XXIII (eds Delgado, M. R., Phelps, E. A. & Robbins, T. W.) (Oxford Univ. Press, New York, 2011).

  34. 34.

    Spiegelhalter, D., Best, N., Carlin, B. & Van der Linde, A. Bayesian measures of model complexity and fit (with discussion). J. R. Stat. Soc. B 64, 583–639 (2002).

  35. 35.

    Gelman, A. et al. Bayesian Data Analysis 3rd edn (CRC Press, Boca Raton, 2013).

  36. 36.

    Myung, I. J. Tutorial on maximum likelihood estimation. J. Math. Psychol. 47, 90–100 (2003).

  37. 37.

    Ahn, W.-Y., Haines, N. & Zhang, L. Revealing neurocomputational mechanisms of reinforcement learning and decision-making with the hbayesdm package. Comput. Psychiatr. 1, 24–57 (2017).

  38. 38.

    Gelman, A., Lee, D. & Guo, J. Stan: A probabilistic programming language for Bayesian inference and optimization. J. Educ. Behav. Stat. 40, 530–543 (2015).

  39. 39.

    Ahn, W.-Y. et al. Decision-making in stimulant and opiate addicts in protracted abstinence: evidence from computational modeling with pure users. Front. Psychol. 5, 849 (2014).

  40. 40.

    Gelman, A. Prior distributions for variance parameters in hierarchical models (comment on article by Browne and Draper). Bayesian. Anal. 1, 515–534 (2006).

  41. 41.

    Gelman, A. & Rubin, D. B. Inference from iterative simulation using multiple sequences. Stat. Sci. 7, 457–472 (1992).

  42. 42.

    Savalia, N. K. et al. Motion-related artifacts in structural brain images revealed with independent estimates of in-scanner head motion. Hum. Brain. Mapp. 38, 472–492 (2017).

  43. 43.

    Cools, R., Clark, L., Owen, A. M. & Robbins, T. W. Defining the neural mechanisms of probabilistic reversal learning using event-related functional magnetic resonance imaging. J. Neurosci. 22, 4563–4567 (2002).

  44. 44.

    Dodds, C. M. et al. Methylphenidate has differential effects on blood oxygenation level-dependent signal related to cognitive subprocesses of reversal learning. J. Neurosci. 28, 5976–5982 (2008).

  45. 45.

    Maldjian, J. A., Laurienti, P. J., Kraft, R. A. & Burdette, J. H. An automated method for neuroanatomic and cytoarchitectonic atlas-based interrogation of fMRI data sets. Neuroimage 19, 1233–1239 (2003).

Download references


The main source of funding for this work was provided by NIMH 105535 R01 grant awarded to I. Harpaz-Rotem and D. Schiller (MPI) and funding provided by the Clinical Neurosciences Division of the National Center for PTSD. Additional support was provided by Klingenstein-Simons Fellowship Award in the Neurosciences to D. Schiller, The Brain and Behavior Research Foundation 23260 to I. Harpaz-Rotem, Chinese NSF grant 31421003 to J. Li, and the Swiss National Science Foundation grant SNF 161077 to P. Homan. The analytic work was supported in part through the computational resources and staff expertise provided by Scientific Computing at the Icahn School of Medicine at Mount Sinai.

Author information

Author notes

  1. These authors contributed equally: Ilan Harpaz-Rotem, Daniela Schiller.


  1. Department of Psychiatry, Icahn School of Medicine at Mount Sinai, New York, NY, USA

    • Philipp Homan
    • , Jingchu Hu
    •  & Daniela Schiller
  2. Departments of Comparative Medicine, Neuroscience and Psychology, Yale University, New Haven, CT, USA

    • Ifat Levy
  3. Department of Psychiatry, Yale University School of Medicine, New Haven, CT, USA

    • Eric Feltham
    • , Charles Gordon
    • , Robert H. Pietrzak
    • , Steven Southwick
    • , John H. Krystal
    •  & Ilan Harpaz-Rotem
  4. US Department of Veterans Affairs National Center for PTSD, Clinical Neurosciences Division, VA Connecticut Healthcare System, West Haven, CT, USA

    • Eric Feltham
    • , Charles Gordon
    • , Robert H. Pietrzak
    • , Steven Southwick
    • , John H. Krystal
    •  & Ilan Harpaz-Rotem
  5. School of Psychological and Cognitive Sciences and Beijing Key Laboratory of Behavior and Mental Health, Peking University, Beijing, China

    • Jian Li
  6. Department of Neuroscience and Friedman Brain Institute, Icahn School of Medicine at Mount Sinai, New York, NY, USA

    • Daniela Schiller


  1. Search for Philipp Homan in:

  2. Search for Ifat Levy in:

  3. Search for Eric Feltham in:

  4. Search for Charles Gordon in:

  5. Search for Jingchu Hu in:

  6. Search for Jian Li in:

  7. Search for Robert H. Pietrzak in:

  8. Search for Steven Southwick in:

  9. Search for John H. Krystal in:

  10. Search for Ilan Harpaz-Rotem in:

  11. Search for Daniela Schiller in:


I.L., I.H.R., and D.S. designed the study. E.F., C.G., I.L., and I.H.R. collected the data. J.H. scored the data. P.H. analyzed the data. I.L., J.L, I.H.R., and D.S. contributed to data analysis. J.H.K., R.H.P., and S.S. contributed to the interpretation of the results. P.H., I.L., I.H.R., and D.S. wrote the first draft of the manuscript. All authors contributed to the final version of the manuscript.

Competing interests

The authors report no competing interests.

Corresponding authors

Correspondence to Ilan Harpaz-Rotem or Daniela Schiller.

Integrated supplementary information

  1. Supplementary Figure 1 Distributions of psychopathology in the study sample show the wide range of symptoms in participants, from completely healthy to highly affected by PTSD.

    ASI, Anxiety Sensitivity Index; BDI, Beck Depression Inventory; CAPS, Clinician-Administered PTSD Scale; CES, combat exposure score; STAIS, State Anxiety subscale of the Spielberger State-Trait Anxiety Inventory.

  2. Supplementary Figure 2 Hierarchical Bayesian model recovers simulated learning parameters.

    To validate our Hierarchical Bayesian hybrid model before testing it in the clinical dataset, we simulated skin conductance data in 20 participants. We estimated the initial associability α0, κ, the normalization factor, and η, the prediction-error weight. a. Simulated data in 20 subjects. b. Correlations between estimated and simulated values. The Bayesian model successfully recovered the simulated values. Pearson correlation coefficients with two-tailed significance tests are shown. Error shadings correspond to standard errors. c. Estimated parameters with 90% credible intervals and simulated parameters. All simulated parameters were within the 90% credible intervals of the estimated parameters, indicating successful parameter recovery. ***, P < 0.001.

  3. Supplementary Figure 3 Average skin conductance response across subjects and the best-fit associability trace for all trials without a shock.

    As there were two experimental orders, time courses are displayed for each order separately (N = 54 participants). All traces were z-transformed and mean-centered for displaying purpose.

  4. Supplementary Figure 4 Model comparison using maximum likelihood estimation (MLE) and the relationship of model fits to PTSD symptoms.

    a. MLE analysis is consistent with the Bayesian analysis and a previous study. In line with a previous study,8 the hybrid model with the lowest Bayesian Information Criterion (BIC) was the hybrid (α + V) model, outperforming the other hybrid models as well as the Rescorla–Wagner (RW) model. The asterisk indicates the winning model. b. No evidence that model fits interacted with the symptom levels of PTSD. The correlation between model fits and symptoms was essentially flat for each of the 4 models. Error shadings correspond to standard errors. PTSD, posttraumatic stress disorder. CAPS, Clinician-Administered PTSD Scale; RW, Rescorla–Wagner model.

  5. Supplementary Figure 5 Hierarchical Bayesian model recovers simulated learning parameters of a Rescorla–Wagner model.

    To validate our Hierarchical Bayesian model before testing it in the clinical dataset, we simulated skin conductance data in 30 participants. We estimated the learning rate α (the free parameter in the Rescorla–Wagner model) using our Hierarchical Bayesian model coded in Stan and also compared it to a Maximum Likelihood Estimation (MLE) using the non-linear optimizer fmincon in MATLAB. The Bayesian model successfully recovered the simulated values.

  6. Supplementary Figure 6 Average skin conductance response across subjects and the best-fit expected value trace for all trials without a shock using a Rescorla-Wagner model.

    As there were two experimental orders, time courses are displayed for each order separately (N = 54 participants). All traces were z-transformed and mean-centered for displaying purpose.

  7. Supplementary Figure 7 Amygdala structure and value computation contribute to PTSD symptoms using a Rescorla–Wagner model.

    a. Regions of interest used in the computational imaging analysis. The amygdala (red) was defined functionally, using the contrast of conditioned stimulus vs. baseline. b-d. Amygdala volume and value-dependent neural activity independently contribute to PTSD symptoms. Partial correlations are shown (N = 54 participants) with Pearson correlation coefficients and two-tailed significance tests. Right amygdala volume and activity as well as left amygdala activity correlated negatively with the PTSD symptoms as measured with the CAPS. Regressions were adjusted for learning rate, age, gender, head movement, and total intracranial volume. Error shadings correspond to standard errors. CAPS, Clinician-Administered PTSD Scale; adj., adjusted; **, P < 0.01; *, P < 0.05.

  8. Supplementary Figure 8 Individual amygdala regions of interest for two illustrative participants.

    To ensure accurate appearance in the published version, please use the Symbol font for all symbols and Greek letters.Brain slices are shown at x = 24, y = −4, z = −16 in the Montreal Neurological Institute (MNI) coordinate system. using the neurological convention (that is, left is left).

  9. Supplementary Figure 9 Regression of prediction error (δ) and associability (α) shows that the two learning components were not significantly correlated.

    Data for both experimental orders (A and B) is shown (N = 69 trials) together with Pearson correlation coefficients and two-tailed significance tests. Error shadings correspond to standard errors.

Supplementary information

  1. Supplementary Text and Figures

    Supplementary Figures 1–9 and Supplementary Table 1

  2. Reporting Summary

About this article

Publication history