Layered fMRI of prediction error related activity in early auditory cortices

Poster No:


Submission Type:

Abstract Submission 


Jakob Heinzle1, Lars Kasper1, Katharina Wellstein1, Johanna Bayer2, Frederike Petzschner3, Ines Pereira3, Matthias Müller-Schrader3, Maria Engel4, Klaas Pruessmann4, Klaas Enno Stephan1


1Translational Neuromodeling Unit, University of Zurich & ETH Zurich, Zurich, Zurich, 2The University of Melbourne, Melbourne, Victoria, 3Translational Neuromodeling Unit, University of Zurich and ETH Zurich, Zurich, Zurich, 4Institute for Biomedical Engineering, ETH Zurich and University of Zurich, Zurich, Zurich

First Author:

Jakob Heinzle  
Translational Neuromodeling Unit, University of Zurich & ETH Zurich
Zurich, Zurich


Lars Kasper  
Translational Neuromodeling Unit, University of Zurich & ETH Zurich
Zurich, Zurich
Katharina Wellstein  
Translational Neuromodeling Unit, University of Zurich & ETH Zurich
Zurich, Zurich
Johanna Bayer  
The University of Melbourne
Melbourne, Victoria
Frederike Petzschner  
Translational Neuromodeling Unit, University of Zurich and ETH Zurich
Zurich, Zurich
Ines Pereira  
Translational Neuromodeling Unit, University of Zurich and ETH Zurich
Zurich, Zurich
Matthias Müller-Schrader  
Translational Neuromodeling Unit, University of Zurich and ETH Zurich
Zurich, Zurich
Maria Engel  
Institute for Biomedical Engineering, ETH Zurich and University of Zurich
Zurich, Zurich
Klaas Pruessmann  
Institute for Biomedical Engineering, ETH Zurich and University of Zurich
Zurich, Zurich
Klaas Enno Stephan  
Translational Neuromodeling Unit, University of Zurich & ETH Zurich
Zurich, Zurich


Proposed implementations of predictive coding in cortical hierarchies postulate predominant processing of prediction errors (PEs) in upper and predictions in lower layers (Bastos et al., 2012). Testing this hypothesis in humans requires fMRI with sufficiently high resolution to distinguish supra- and infragranular layers (Stephan et al., 2019). Here, we measured laminar fMRI during an auditory mismatch paradigm and investigated differences in profiles of explained variance between PE and prediction activity. In addition, we directly investigated the relative strengths of upper and lower layer activations.


We acquired high resolution fMRI from 24 participants on a 7 Tesla Philips scanner using a spiral readout (0.9mm isotropic resolution, TR=3.12s). Every 1.1s, a visual pacing stimulus was followed by the presence or absence (omission) of a tone in a roving fashion. An extended MR signal model combined with magnetic field monitoring of the spiral readouts was used (Kasper et al., 2019). In addition, we acquired T1 weighted (0.8mm isotropic) and multi-echo gradient-echo (ME-GRE, 1 mm isotropic) anatomical scans. T1 images were aligned to the IR and segmented with Freesurfer. Upper and lower layers were delineated ( at 25% and 75% cortical depth, respectively. Functional images were slice time corrected, realigned and coregistered to the ME-GRE image. For each voxel, we calculated the distance to the two layers as well as its cortical depth.
Statistical analysis was based on a general linear model that included a regressor modelling the onset of each tone, four modulatory regressors (PEs and predictions for tones and omissions, respectively), and physiological nuisance regressors (Kasper et al., 2017). Modulators were calculated using the hierarchical Gaussian filter (Mathys et al., 2011). In order to assess the importance of individual regressors we computed voxel-wise extra sum of squares (ESS).
The layered analysis focused on the modulatory regressors and was restricted to areas that showed an effect of tone within auditory areas A1 and superior temporal gyrus (STG). Two approaches were used. First, distinguishing only upper and lower layers, we computed a distance-weighted average ESS of all voxels within 0.5 mm of the respective surface. For tones and omissions, we then compared the activation of the lower layer relative to the upper layer (in terms of ESS) between PEs and predictions. Second, we computed profiles across the cortical depth by averaging ESS from all voxels within ten different cortical depth ranges. We normalized profiles by dividing by the ESS in a bin just outside the pial surface. We used permutation tests and corrected for multiple comparisons (Maris and Oostenveld, 2007), with a significance level of α=0.05. All analyses were performed using SPM12 and Matlab (MathWorks, Natick, MA).


Fiugre 1 summarizes the results. When comparing only relative activations of two layers, lower layer activation by predictions was larger than by PEs for both tones and omissions in A1. In normalized depth profiles of ESS, significant differences between predictions and PEs were observed in lower cortical depths in A1 for both tones and omissions. In STG, only tones showed this behavior.
Supporting Image: FigureAbstract.png
   ·Figure 1: Results


The most salient finding, replicated in both analyses, is that predictions show less attenuation from the pial surface towards the white matter than PEs. These profiles are in line with the notion that PE activity is relatively more pronounced in upper layers and prediction activity in lower layers. Importantly, in our analyses, known blood draining effects impact all quantities in the same way, thus our interpretations are unlikely to be affected by hemodynamic confounds. Future work will include detailed modelling of blood draining effects (Havlicek and Uludağ, 2020; Heinzle et al., 2016) to corroborate this finding.

Modeling and Analysis Methods:

Activation (eg. BOLD task-fMRI) 2

Novel Imaging Acquisition Methods:


Perception, Attention and Motor Behavior:

Perception: Auditory/ Vestibular


Cortical Layers

1|2Indicates the priority used for review

My abstract is being submitted as a Software Demonstration.


Please indicate below if your study was a "resting state" or "task-activation” study.


Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Healthy subjects

Was any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.


Was any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.

Not applicable

Please indicate which methods were used in your research:

Functional MRI
Structural MRI

For human MRI, what field strength scanner do you use?


Which processing packages did you use for your study?

Free Surfer

Provide references using author date format

Bastos, A.M., Usrey, W.M., Adams, R.A., Mangun, G.R., Fries, P., Friston, K.J., 2012. Canonical microcircuits for predictive coding. Neuron 76, 695–711.
Havlicek, M., Uludağ, K., 2020. A dynamical model of the laminar BOLD response. NeuroImage 204, 116209.
Heinzle, J., Koopmans, P.J., den Ouden, H.E., Raman, S., Stephan, K.E., 2016. A hemodynamic model for layered BOLD signals. NeuroImage 125, 556–70.
Kasper, L., Bollmann, S., Diaconescu, A.O., Hutton, C., Heinzle, J., Iglesias, S., Hauser, T.U., Sebold, M., Manjaly, Z.M., Pruessmann, K.P., Stephan, K.E., 2017. The PhysIO Toolbox for Modeling Physiological Noise in fMRI Data. Journal of neuroscience methods 276, 56–72.
Kasper, L., Engel, M., Heinzle, J., Mueller-Schrader, M., Reber, J., Schmid, T., Barmet, C., Wilm, B.J., Stephan, K.E., Pruessmann, K.P., 2019. Advances in Spiral fMRI: A High-resolution Study with Single-shot Acquisition. bioRxiv.
Mathys, C., Daunizeau, J., Friston, K.J., Stephan, K.E., 2011. A Bayesian foundation for individual learning under uncertainty. Frontiers in human neuroscience 5.
Stephan, K.E., Petzschner, F.H., Kasper, L., Bayer, J., Wellstein, K.V., Stefanics, G., Pruessmann, K.P., Heinzle, J., 2019. Laminar fMRI and computational theories of brain function. NeuroImage 197, 699–706.