Experimental design.

Subjects participated in two fMRI sessions with 15 scans each: encoding, recognition, and recall, repeated five times. The encoding and recall scans included 24 trials each, corresponding to 24 distinct objects, in random order. Recognition scans contained the same 24 objects, with an additional 24 new objects (lures), shown in random order. In encoding trials, subjects were briefly shown an object in one of four locations at 2° eccentricity (45°, 135°, 225°, or 315° polar angle) and asked to indicate whether they were larger or smaller than a shoebox in real life. In recognition trials, subjects were shown an object at fixation and were asked to indicate whether it was old or new. In recall trials, subjects were shown an old object at fixation, and were asked to indicate its original location.

Reanalysis excluding experimenter subjects.

Behavioral performance on recognition and recall tasks.

A) Subject-level performance on the recognition task. Individual subjects are plotted as circles for hit rate (filled) and false alarm rate (unfilled). Black horizontal lines indicate means across subjects. The dotted line indicates chance performance. B) Subject-level performance on the recall task. C) Recognition performance across blocks. Data points are the means across subjects for hit rate (filled) and false alarms (unfilled). Error bars indicate 95% CI obtained from 10,000 bootstraps across subjects. D) Recall performance across blocks. The code and data to generate this and all subsequent figures can be found at https://osf.io/a9hkg/.

Single-shot memory responses in visual cortex are tuned to encoded location.

A) Mapping BOLD responses to polar angle activation profiles. A stimulus evokes a BOLD response in visual cortex. This activity is projected to visual space and binned by angular distance from target, producing a polar angle activation profile. This example profile is from V3 during the encoding task. B) Polar angle activation profiles in V1-V3 across the three tasks. Shaded regions indicate 68% confidence intervals across subjects. C-F) Estimates of tuning quality for the three tasks. Error bars indicate 68% confidence intervals from bootstrapping across subjects. For data plotted separately by visual area (V1, V2, V3) or by individual participant, see Supplemental Figure 3-1 and Figure 4, respectively. C) Distance between peak location and object source location, computed as absolute deviation of the peak of the polar angle activation profile. D) Response amplitudes (peak to trough). The y-axes have different scales for encoding and retrieval. E) Tuning width (full-width at half the maximum amplitude). F) Angular fidelity, a summary metric that depends on the peak location, width, and signal-to-noise ratio in the response profiles

Spatial tuning during encoding, recognition and recall for V1, V2, and V3.

Polar angle activation profiles during encoding, recognition, and recall for each visual map V1, V2, and V3. Spatial tuning is evident within each visual map. V1 encoding fidelity was at 0.65 [95% CI 0.58, 0.74], recognition was 0.31 [0.1, 0.42], and recall was 0.26 [0.13, 0.33]. V2 encoding fidelity was at 0.63 [0.6, 0.68], recognition was 0.27 [-0.05, 0.48], and recall was 0.17 [-0.05, 0.4]. V3 encoding fidelity was at 0.71 [0.63, 0.73], recognition was 0.35 [0.14, 0.5], and recall was 0.32 [0.1, 0.41].

Recognition memory response by block number in V1.

A) Polar angle activation profile for V1 during the first recognition block, restricted to trials with subsequent accurate recall (i.e. location) memory. The first recognition block differs from the others in that participants are naive to the recall task in the first recognition block. Even so, the recognition memory response showed robust tuning near the original object’s location, with a fidelity of 0.21 [95 CI -0.02, 0.42]. B) Fidelity of the recognition memory response across blocks. Shaded region denotes block 1 plotted in panel A. Error bars are 68% confidence intervals bootstrapped across subjects. The fidelity of the recognition memory response does not increase in subsequent blocks, which shows that knowing about the recall task does not improve the recognition memory response.

Gaze position and saccades in during each task’s stimulus display.

A) Density plot of average gaze position within each trial’s 2-s stimulus display window. Each object’s source location was at 2° eccentricity. Shaded inset: Group-level mean gaze position for different stimulus source locations revealed only minor deviations from fixation point across all tasks (<= 0.1 deg) (n = 13). Error bars represent the 95% confidence intervals of the mean. White numbers indicate source location (1: 45°, 2: -45°, 3: -135°, 4: 135°). The average calibration accuracy of the eye tracker across participants (who had at least two validation runs) was 0.64° [95 CI 0.49°, 1.04°], which is consistent with Eyelink’s reported calibration accuracy levels of 0.5°. B) Top row: Density histogram of saccade eccentricity within each trial’s stimulus display window. Saccades were rare (8.7% of sampled frames across participants [68 CI 7.3%, 10.1%]). Less than 10% of saccades landed closer to the target eccentricity than to the fixation dot (shaded region) in both memory retrieval tasks (encoding = 13.6%, recognition = 8.1%, location = 8.8%). Bottom row: Density histogram of saccade polar angle during stimulus display window for saccades that landed closer to the target eccentricity than fixation point, aligned to the source location of the trial’s object. Dashed gray lines mark polar angle location of stimuli (0°). Aligning near-target saccades to their respective trial’s source location reveals a small bias to the source location during the encoding task, but not during either memory retrieval task. These results demonstrate that subjects are maintaining central fixation throughout the experiment, and that the tuned recognition and recall responses are not an artifact caused by saccades biased towards each object’s source locations.

Spatial tuning during encoding, recognition, and recall for individual subjects.

V1-V3 polar angle activation profiles for all subjects (n=20) and tasks. Transparent gray overlay indicates the subject excluded from group-level analyses due to not following response report instructions during the recall task. The second and third row in the first column belong to the two experimenter subjects. Dashed grey lines indicate 0°, i.e., the polar angle location of the object at encoding. Asterisks denote responses with angular fidelity greater than 1 (*) or 2 (**) sd from the mean of the null distribution. The null distribution was computed separately for each subject and each condition by scrambling the assignment of target locations across trials, and then re-computing the fidelity, 1,000 times per participant.

Spatially-tuned memory responses in early visual cortex track recall success.

A) Schematic for grouping remembered/forgotten trials across tasks. Trials in each task (encoding, recognition, and recall) are assigned to the ‘remembered’ group if the object’s location is reported correctly during the recall task. Trials are deemed ‘forgotten’ if location is reported incorrectly during the recall task. B) Polar angle activation profiles during both memory retrieval tasks, for trials where the object’s location was remembered at recall. C) Polar angle activation profiles for trials where the object’s location was forgotten at recall.

Memory responses for remembered vs. forgotten trials in V1, V2, V3.

Polar angle activation profiles for V1, V2, and V3 during both memory retrieval tasks. Trials are grouped according to whether the object location probe was answered correctly (top row) or not (bottom row) during the recall task. Numbers inside the subplots indicate the fidelity of the polar angle activation profile and its 95% confidence interval. The drop in fidelity between remembered and forgotten trials is most evident in V1.

Report-aligned memory responses for forgotten trials for all ROIs.

Report-aligned polar angle activation profiles for V1-V3 and each visual map during both memory retrieval tasks for trials where the object’s location was forgotten. Numbers inside the subplots correspond to the fidelity of the polar angle activation profile and its 95% confidence interval. The tuning of the recognition and recall responses for forgotten trials was low across ROIs when the responses were aligned to the behavioral report.

Spatial tuning in early visual cortex at encoding predicts subsequent recall.

A) Polar angle activation profiles from early visual cortex during encoding, grouped by whether the encoded object’s location was later remembered (blue) or forgotten (gray) during the recall task. Shaded region indicates 68% confidence interval across subjects. B) Amplitude estimates (peak to trough) during encoding for remembered and forgotten groups. Error bars indicate one standard deviation of the amplitude difference between the two groups across bootstraps. C) Differences in BOLD response between remembered and forgotten groups as a function of angular distance to the target. Error bars indicate one standard deviation of the bootstrapped differences between the two groups. For separate plots of V1, V2, and V3, see Supplemental Figure 6-1.

Subsequent memory effect in V1, V2, and V3.

Differences in BOLD response between remembered and forgotten trials as a function of angular distance to the target, for each visual map V1, V2, and V3. Error bars indicate one standard deviation of the bootstrapped differences between the two groups. The dashed grey line indicates equality between groups. V3 shows the largest difference between remembered and forgotten responses during encoding, smallest difference in V1.