Increased activity in superficial and deep layers of human S1 for prediction error
Yinghua Yu1,2,3, Laurentius Huber3, Yuhui Chai3, David C Jangraw3, Arman Khojandi3, Jiajia Yang1,3, and Peter A Bandettini3

1Okayama University, Okayama, Japan, 2The Japan Society for the Promotion of Science, Tokyo, Japan, 3National Institute of Mental Health, Bethesda, MD, United States


Sensory processing in humans is thought to rely on a predictive model of the environment. And these predictions are constantly optimized to minimize future sensory prediction errors. However, the neural microcircuits underlying this prediction error model are still poorly understood. Here, we used an index finger prediction task that consists of sequential finger-stroking in high-resolution (0.71mm) BOLD and VASO fMRI at 7T to investigate how the prediction error activity changes across layers in the human primary somatosensory cortex (S1). We found that prediction error activity is stronger in superficial and deep layers rather than the middle layers of S1.


When humans perceive a sensation, their brains represent the inputs from sensory receptors and continuously update these representations based on expectations. This mechanism of predictive coding is based on the hierarchical processing of prediction errors that are calculated by the predictive feedback and feedforward input1,2. In our previous study3, we presented the first application in humans of high-resolution (0.71mm) layer-dependent fMRI in the technically challenging area of primary somatosensory cortex (S1). We found that sensory input to S1 evoked activity in middle laminae, while mental prediction induced activity in superficial and deep laminae (Figure 1(e)&(f)). However, the precise contributions of specific layers to prediction error processing are not fully understood to date in humans. To explore the layer contributions in prediction error processing in human S1, we acquired high-resolution fMRI4,5 at 7T and sought to identify layer-specific activity in area 3b by using a series of index finger prediction tasks.


A 7T Siemens scanner, equipped with a 32-channel NOVA Medical head coil and SC72 body gradient (Figure 1(a)) was used. Data acquisition procedures were used as described in our previous study3,4,5. The timing of the acquisitions was: TI1/TI2/TR=1100/2845/3490 ms. The coil-combined data consist of interleaved BOLD and blood volume sensitive VASO6 contrasts – obtained as separate yet concomitant time series. These time series are corrected for rigid volume motion and are separated by contrast with an effective temporal resolution of TR = 3.49 s. The nominal resolution was 0.71 mm across cortical layers with 1.8-mm thick slices perpendicular to the postcentral bank of the right central sulcus (Figure 1(b)&(c)). To investigate the cortical layer-dependent brain activations that are reflecting the prediction error, four participants were asked to undergo one or two 16 min fMRI runs. As shown in Figure 2(a), each run consisted of three conditions which were designed to include prediction tasks without error (Match stroking task, MS), prediction with error (non-Match stroking task, nMS) and unpredictable sensory input (Random stroking task, RS). All those tasks alternated between 34 sec “on” vs. 20 sec “off” and each condition was repeated four times. The participant was asked to predict when the left index finger will be stroked in MS and nMS tasks, RS tasks, however, they were asked to simply pay attention to the stimulation without mental prediction involved. An additional digit localizer task was conducted to identify somatotopically organized ROIs of the individual fingers. Laminar analyses were conducted with the open software suite LAYNII https://github.com/layerfMRI/LAYNII.


Prediction task-induced fMRI signal change in the area 3b was found in all participants. As shown in Figure 2(b), layer-dependent activity for BOLD-fMRI modulations across tasks could be detected in a representative participant’s individual activation maps with and without smoothing along the cortical layers. Specifically, MS and nMS tasks evoked strong activations across all laminae in area 3b, however, the RS task evoked activations in the middle input layer only. Averaged profiles of layer-dependent BOLD and VASO responses for the three task conditions are shown in Figure 3(a). The activity in superficial and deep layers showed clear differences across the three conditions. Consistent to our previous finding3, the contrast of MS vs. RS showed that mental prediction evokes two peaks of activation in superficial and deep layers. Surprisingly, prediction errors (nMS vs. RS in Figure 3(b) black line), caused stronger activation peaks in superficial and deep layers than those of prediction without error (MS vs.RS in Figure 3(b) pink line).


In the present study, our main finding is that both superficial and deep layers of area 3b play a crucial role in the prediction errors related processing. In line with the predictive coding principle1,2,7, these stronger activations in superficial and deep layers are related to the computation and minimization of prediction errors engaged in somatosensory processing. Specifically, the prediction error signals might originate in the superficial layers and then project to the deep layers for further error correction processing. Our findings provide insights into how these laminar circuits represent the prediction error details, and we regard these finding as an important step towards the understanding of predictive coding processing dynamics.


We used sub-millimeter BOLD and blood-volume-sensitive (VASO) fMRI3,4,5 at 7T to demonstrate that the superficial and deep layers of human area 3b play a crucial role in the prediction error-related processing such as error computation and minimization. This laminar specificity was directly visible on functional MRI maps during task-dependent activity changes. Furthermore, the use of VASO fMRI, which is specific and sensitive enough to reveal functional laminar activity, allowed us to focus on directional activation patterns at different cortical layers.


We thank Benedikt Poser and Dimo Ivanov for the 3D-EPI readout that is used in the VASO. We thank Kenny Chung for administrative radiographic assistance of the human volunteer scanning. This research is supported by the NIMH Intramural Research Program (#ZIA-MH002783), JSPS KAKENHI Grant Numbers JP17J40084, JP18K15339, JP18H05009, JP17K18855 and Japan-U.S. Science and Technology Cooperation Program (Brain Research). The study was approved under NIH Combined Neuroscience Institutional Review Board protocol #93-M-0170 (ClinicalTrials.gov identifier: NCT00001360).


1. Friston KJ. The free-energy principle: A unified brain theory? Nat Rev Neurosci. 2010, 11:127–138

2. ´╗┐Barrett LF, Simmons WK. Interoceptive predictions in the brain. Nat Rev Neurosci. 2015, 16:419–429

3. Yu Y, Huber L, Jangraw DC, Molfese PJ, Hall A, Handwerker DA, Yang J, Bandettini PA. Depth-dependent functional mapping of mental prediction in human somatosensory cortex. Proc Intl Soc Mag Reson Med. 2018, 26:393

4. Huber L, Handwerker DA, Jangraw DC, Chen G, Hall A, Stüber C, Gonzalez-Castillo J, Ivanov D, Marrett S, Guidi M, Goense J, Poser BA, Bandettini PA. High-Resolution CBV-fMRI Allows Mapping of Laminar Activity and Connectivity of Cortical Input and Output in Human M1. Neuron 2017, 1253–1263

5. Huber L, Ivanov D, Handwerker DA, Marrett S, Guidi M, Uluda─č K, Bandettini PA, Poser BA. Techniques for blood volume fMRI with VASO: From low-resolution mapping towards sub-millimeter layer-dependent applications. Neuroimage 2018, 164:131–143

6. Lu H, Golay X, Pekar JJ, van Zijl PCM. Functional magnetic resonance imaging based on changes in vascular space occupancy. Magn. Reson. Med. 2003, 50: 263–274

7. Bastos AM, Usrey WM, Adams RA, Mangun GR, Fries P, Friston KJ. Canonical Microcircuits for Predictive Coding. Neuron 2012, 76:695–711


Figure 1. Acquisition methods and expected fMRI activity across layers. Slice-selective slab-inversion VASO was used on a 7T scanner (a), equipped with a 32-channel RF coil and a SC72 body gradient coil. (b) (c) The nominal resolution was 0.71 mm across cortical layers with 1.8-mm thick slices perpendicular to the postcentral bank of the right central sulcus. (d) Finger somatotopy in area 3b for one participant. Expected layer-dependent circuitry (e) and fMRI activity (f) from our previous study3.

Figure 2. Illustration of the temporal match and non-match finger stroking tasks and the corresponding activation maps of a representative participant. (a) The three different conditions for each functional task are shown. The prediction error in the nMS condition was introduced by modulations the stoking timing. The participants had learned to be stroked in a rhythmic fashion. However, for the nMS condition an additional delay was introduced, such that the stroking occurred later than expected. (b) The first row represents unsmoothed activation maps of three different conditions of one participant. For visualization, the second row represents activation maps with smoothing in each lamina.

Figure 3. Cortical profiles of BOLD and VASO activity changes in the ROI of the index finger in area 3b, averaged across participants. (a) The three tasks resulted in modulated cortical activity profiles. (b) Prediction error modulated activity is dominant in superficial laminae and deep laminae (pink and black lines differ most at positions of red arrows).

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)