Paradigm.

A) Example trial of the additional singleton visual search task. Trials started with a placeholder display (800 ms duration) signifying trial onset, followed by a fixed inter-stimulus interval of 400 ms. On 75% of trials a search trial appeared next for 1500 ms. On 25% of trials, an omission trial (identical to the ISI display) was shown instead. Trials ended with a variable ITI of ∼4000 ms. In the example search trial, the target is the horizontal green bar in the right upper corner, because it has a different orientation compared to the other seven bar stimuli. Because the target contains no black dot in the center, the correct response is ‘no’. The highly salient distractor is the orange vertical bar. The central fixation cross and octagonal outline surrounding the possible search area were presented throughout the experiment. B) Three example color inversions of the location localizer, sampling the upper left location. The location localizer cross had the same size and location as the two possible bar stimuli (horizontal and vertical) during the search trial overlaid over each other, thus sampling neural populations responsive to this location. The cross flickered at 4 HZ for 12 seconds for each location of interest, sampling each location 8 times per run.

Behavioral facilitation by distractor suppression.

A) Reaction times (RT in millisecond; ordinate) were faster when distractor stimuli appeared at the HPDL compared to NL-near or NL-far locations. RTs were fastest when no distractor was present. B) Response accuracy (in percent; ordinate) for each distractor location. Asterisks indicate statistically significant post hoc tests: * p < 0.05, *** p < 0.001.

Illustration of the analysis rationale and hypotheses.

A) ROI analysis procedure. During an independent location localizer task (left) checkerboard cross patterns (flickering black and white at 4 HZ) were presented at the locations corresponding to the bar locations during the search task (right). Using BOLD activations in EVC from this localizer, location specific ROI masks were created for the four locations of interest (HPDL, two NL-near, NL-far). The masks were then applied to the additional singleton task and the contrast parameter estimates (BOLD) during search trials extracted in a stimulus and location specific manner. To illustrate, in the example above we assume that the HPDL was at the upper left location (red circle; determined by the statistical regularities throughout the search task). The example search trial contained a neutral stimulus at the HPDL (red circle), a salient distractor at the left NL-near (blue circle), a neutral stimulus at the right NL-near (dark blue circle) and a target stimulus at the NL-far (green circle) location. Therefore, the data provided by this trial was a neutral stimulus at HPDL (red arrow), a distractor at NL-near (blue arrow), a neutral stimulus at the other NL-near (not depicted), and a target at NL-far (not depicted). Therefore, each trial provided multiple location specific datapoints. Specifically, data for each stimulus type and location combination were first estimated across trials and then extracted using the ROI based approach. Data across the two NL-near locations were averaged. For further details see Materials and Methods: Statistical Analysis and Region of interest (ROI) definition. The same ROI analysis was performed for omission trials, except that by design, omission trials did not contain stimuli, hence resulting in only location specific activation data points. B) Potential outcomes for search trials. We distinguish between two factors modulating BOLD responses during search trials. First, we asked whether modulations in EVC are spatially specific. Illustrated on the left is a spatially unspecific modulation, affecting neural populations with receptive fields at all three locations equally. The middle panel depicts a spatially specific modulation with a gradient of increasing suppression the closer a location is to the HPDL. Second, we ask whether BOLD modulations are stimulus specific, that is selectively suppressing only distractor stimuli, but not target and neutral stimuli (right panel). C) Additionally, we distinguished between reactive compared to proactive spatial modulations by contrasting BOLD during omission trials. Reactive modulations (i.e., following search display onset) result in no spatially specific effects during omission trials (left panel), because no search display was shown. In contrast, proactive suppression yields spatially specific BOLD modulations during omission trials due to the deployment of spatial priority maps by anticipated search (right panel).

Distractor suppression in early visual cortex.

A) fMRI BOLD responses (ordinate) during search trials, split into stimulus types (abscissa). Color denotes locations based on distractor contingencies with red = high-probability distractor location (HPDL), blue = neutral locations near the HPDL (NL-near), green = neutral locations far from the HPDL (NL-far; diagonally opposite from the HPDL). BOLD responses were systematically suppressed for all stimuli occurring at the HPDL and NL-near compared to NL-far. B) Corresponding results for omission trials. Neural populations with receptive fields at the HPDL and NL-near locations were suppressed compared to those with receptive fields at the NL-far location. Note that BOLD responses are overall close to zero, which is expected given that no display was shown at this time. Asterisks indicate statistically significant pairwise comparisons within stimulus types: * p < 0.05, ** p < 0.01, *** p < 0.001.

Ruling out explicit attentional strategies: Distractor suppression in a subsample of participants with incorrect HPDL choices in the questionnaire.

Results for search (A) and omission (B) trials were highly similar to the main results using the full sample. Significant suppression of BOLD responses at both the HPDL and NL-near locations compared to the NL-far location were evident during search and omission trials in the subsample with incorrect responses for the HPDL location on the questionnaire probing explicit knowledge of the distractor contingencies. Asterisks indicate statistically significant pairwise comparisons within stimulus types: * p < 0.05, ** p < 0.01.

No behavioral prioritization of targets at NL-far.

Behavioral data in a target contingent analysis. To avoid confounds by distractor location, only distractor absent trials were included in the target contingent analysis. A) Reaction times (RT in millisecond; ordinate) did not significantly differ when targets appeared at the NL-near compared to the NL-far location. B) There were no response accuracy (in percent; ordinate) differences when targets were presented at the HPDL or NL-near locations compared to NL-far. Both RT and response accuracy results contradict a strategic prioritization of the NL-far location.

fMRI results generalize across ROI mask sizes.

To ensure that our results were not dependent on the exact number of voxels in the ROI masks, we repeated the main fMRI analysis with varying mask sizes. For simplicity, and because the primary ROI results suggested that there were no major differences between stimulus types, we here collapsed across stimuli by averaging distractor, target, and neutral stimuli. A) Results for search trials and B) omission trails were largely invariant to ROI size up to very large masks of 200 or more voxel per location (i.e., 800 or more voxels total, 6400 mm3). Asterisks indicate statistically significant simple main effects of location. * p < 0.05, ** p < 0.01.

fMRI results did not depend on location specific normalization of BOLD responses.

The HPDL differed between participants to avoid systematic effects of BOLD response differences driven by the physical location or hemodynamic differences for specific locations beyond the statistical regularities. However, within each participant the HPDL was constant, thus overall response magnitude differences per location may still affect the BOLD results. Therefore, in our main ROI analysis we normalized fMRI BOLD responses for each location separately using independent localizer data. Specifically, we divided the BOLD response during the main task by the BOLD response during the localizer run for each location separately, thereby correcting for any location specific differences in overall BOLD responses. Importantly, during the localizer no statistical contingencies were present and only one checkerboard stimulus was shown at a time. We believe that this approach is preferrable, however for transparency we also report the non-normalized results here. A) Search trials. While results were overall less robust, which was expected given the additional noise added by location specific BOLD differences, they qualitatively matched the normalized results. BOLD differences due to statistical learning were evident for NL-far compared to the other locations for target stimuli and NL-far compared to NL-near for distractor and neutral stimuli. B) For omission trials results qualitatively fit the main ROI analysis results, with a main effect of location and a reliable difference between NL-far and NL-near locations. In sum, non-normalized results further support our conclusions. Asterisks indicate statistically significant pairwise comparisons within stimulus types. * p < 0.05, ** p < 0.01, *** p < 0.001.

ROI location masks generalize across localizer runs.

Depicted are results from a cross-validation analysis, confirming the stimulus location selectivity of our EVC ROI masks. Number of voxels are shown on the x axis, while the y axis depicts the averaged contrast parameter estimates of the location contrasts (average all combinations of stimulation of one location vs the other three locations) from the location localizer. For this cross-validation analysis masks were defined using data from the first localizer run and tested on the second localizer run (and vice versa). Location selectivity contrast parameter estimates were obtained for each participant separately and then averaged across subjects. Thus, values larger than zero indicate successful generalization of the location selectivity masks from one run to the other. All one-sample tests, contrasting stimulus location selectivity against zero (no selectivity), were p < 0.001, demonstrating that EVC ROI definition was reliable and successful. Error bars denote the SEM. Note that up to 400 voxels per location mask zero voxels overlapped between location masks, further bolstering mask selectivity.

Results of planned pairwise tests contrasting fMRI BOLD responses during search trials.

Contrasted are the three stimulus locations (HPDL, NL-near, NL-far) for each stimulus type (distractor, target, neutral stimulus) separately. Reported are paired t-tests or Wilcoxon signed-rank tests results as appropriate with associated effect sizes (Cohen’s d for t-tests and matched rank biserial correlation for Wilcoxon signed-rank tests). P values are uncorrected.