Repetition suppression (RS) reflects a neural attenuation during repeated stimulation. We used fMRI and the subsequent memory paradigm to test the predictive coding hypothesis for RS during visual memory processing by investigating the interaction between RS and differences due to memory in category-selective cortex (FFA, pSTS, PPA, and RSC). Fifty-six participants encoded face and house stimuli twice, followed by an immediate and delayed (48 h) recognition memory assessment. Linear Mixed Model analyses with repetition, subsequent recognition performance, and their interaction as fixed effects revealed that absolute RS during encoding interacts with probability of future remembrance in face-selective cortex. This effect was not observed for relative RS, i.e. when controlled for adapter-response. The findings also reveal an association between adapter response and RS, both for short and long term (48h) intervals, after controlling for the mathematical dependence between both measures. These combined findings are challenging for predictive coding models of visual memory and are more compatible with adapter-related and familiarity accounts.

Subjects were recruited via advertisements for participation in an fMRI memory experiment.

56 healthy subjects participated in our study. The final sample for the main analyses consisted of fifty-four subjects [13 males (24 %); mean age ± SD = 34 ± 11 years, range 21-64]. 

Sequence & imaging parameters: Brain imaging was performed on a 3T Siemens Achieva scanner, using a 32-channel head coil. Acquisition parameters for 45 participants consisted of a high-resolution T1-weighted anatomical image (voxel size: 0.98 x 0.98 x 1.20 mm3) using a 3D turbo field echo sequence (TR:9.6 ms; TE:4.6 ms; matrix size:256 x 256; 182 slices); a T2*-weighted GE-EPI sequence with the following parameters: TR: 2000 ms; TE: 30 ms; matrix size: 80 x 78; FOV: 230 mm; flip angle: 90˚; slice thickness: 4 mm; no gap; axial slices: 38. For the other 9 participants, a similar high-resolution T1-weighted anatomical image was acquired (voxel size: 1.10 x 1.10 x 1.10 mm3) using a 3D turbo field echo sequence (TR:6.9 ms; TE:3.2 ms; matrix size:256 x 256; 208 slices) and a T2*-weighted GE-EPI sequence with the following parameters: TR: 2000 ms; TE: 30 ms; matrix size: 80 x 78; FOV: 230 mm; flip angle: 90˚; slice thickness: 4 mm; no gap; axial slices: 36. Scan acquisition setting was included as a nuisance variable in all brain imaging analyses.

Preprocessing software: Imaging data were analyzed using BrainVoyager 21.4. Pre-processing of functional data consisted of slice scan time correction, temporal high-pass filtering to remove low-frequency drifts, realignment to the first image to compensate for head motion, and spatial smoothing with a Gaussian filter of 4mm FWHM. Functional data were co-registered with the anatomical images and normalized into Talairach coordinate space. 

Model type and settings: At first level, the statistical analysis was based on the general linear model (GLM) with repetition (no, yes), category (face, house), and subsequent memory performance (Not, probably yes, and definitely yes) as factors. The ‘definitely not’ and ‘probably not’ conditions were pooled as 25 participants did not use both categories during their experiment in one of both categories. Null-events were modelled explicitly. At second level, a random effects GLM was performed.

Linear Mixed Models were estimated with the beta-values as dependent variable with repetition (2 levels: adapter and test), memory performance (3 levels: forgotten, probably remembered or definitely remembered), and repetition x memory performance as fixed effects using an unstructured variance-covariance matrix (based on a Likelihood test). These analyses were performed for every ROI separately. We applied Bonferroni-correction for multiple comparisons (‘forgotten’ vs ‘probably remembered’, ‘forgotten’ vs ‘definitely remembered’, and ‘probably remembered’ vs ‘definitely remembered’). We only focused on within-performance category differences. In addition to the main effects of performance and repetition, we performed post-hoc analyses to study the interaction between performance and region and the interaction between repetition and region. 

One participant was excluded due to technical failure during fMRI acquisition and one participant was excluded due to indication of pathology. 

The statistical threshold for ROI-definition at subject level was set at P_height<.005. Two participants did not show any significant results for the right FFA, five for the left FFA, one for the right pSTS, two for the left pSTS, and four for the left RSC. These participants were excluded from the respective analyses.